The Influence of Street-Scene Emotional Perception on Commercial Satisfaction: A Study Based on Random Forest Model and SHAP Algorithm

The Influence of Street-Scene Emotional Perception on Commercial Satisfaction: A Study Based on Random Forest Model and SHAP Algorithm | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article The Influence of Street-Scene Emotional Perception on Commercial Satisfaction: A Study Based on Random Forest Model and SHAP Algorithm Yiwei Mo, Chao Luo, Youpeng Lu, Qi Wu, Jingjing Wang, Zixiang Feng This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8017731/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract In cross-disciplinary research on urban planning and service marketing, the street landscape is seen as an extended service environment. However, most studies focus on the link between a single business type and one emotional dimension. This leaves the complex mechanisms between streetscape emotional perception and commercial satisfaction unrevealed. This study combines street-view imagery, Dianping point-of-interest (POI) data, and online ratings to build a dataset. It covers four main commercial categories (Catering Services, Retail and Trade Locations (RTL), In-store Service Sectors, Leisure and Experience Venues) and six emotional perceptions (Safety, Aesthetics, Wealth, Vibrancy, Depression, Boring). The study uses Random Forest (RF) regression and SHapley Additive exPlanations (SHAP) to examine how streetscape emotions influence businesses. Results show catering services are mostly affected by the Depression and Vibrancy (Dpr–Vib) link. In-store services and leisure venues are shaped by the synergy of Aesthetics and Wealth (Aes–Wth). Retail trade reveals a tension between Aesthetics and Depression. The study proposes a business-type-specific "emotion-customized" spatial intervention framework to guide urban renewal and management of mixed-use streets. Earth and environmental sciences/Environmental social sciences Social science/Environmental studies Scientific community and society/Geography Social science/Geography Street-level Emotion Perception Service Landscape Theory Commercial Satisfaction Business Format Heterogeneity Random Forest Emotion-Customized Spatial Intervention Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1 Introduction Urban streets are not only transportation networks connecting different functional zones but also constitute primary public spaces for social life through the spatial integration of roads, buildings, and associated service facilities (Mehta, 2013 ). Under the combined influence of streetscape characteristics and commercial functions, necessary, optional, and social activities such as walking, shopping, leisure, and social interaction naturally intertwine and transform, collectively shaping the attractiveness and vibrancy of urban streets (Istrate, 2025 ; Whyte, 1980 ). Grounded in servicescape theory Bitner, an increasing number of interdisciplinary studies have demonstrated the associations between streetscape features, emotional perception, and commercial satisfaction (Han et al., 2025 ; Koo et al., 2023 ). However, several limitations remain. First, many studies tend to overemphasize safety perception or focus on a single business category (e.g., catering services), often simplifying "street commercial satisfaction" into a homogeneous concept while overlooking its differentiated functional and emotional mechanisms. Empirical evidence, however, suggests that different business types—such as catering, retail, cultural, and entertainment sectors (—exhibit varying sensitivities to emotional cues (Correia & Kozak, 2016 ; Meng et al., 2017 ), most previous research has primarily relied on linear modeling approaches, which may fail to capture potential nonlinear or threshold effects, as well as the interactive mechanisms between emotional perception and commercial functionality. Such methodological limitations may restrict the explanatory power and predictive robustness of existing models (Li et al., 2024 ; Mittal & Kamakura, 2009 ) . To address these limitations, this study takes Wuhan, China, as a case to construct a multi-category analytical framework linking streetscape emotion and commercial satisfaction. Specifically, the study:(1) applied a deep learning model trained on the MIT Place Pulse dataset to extract six perceptual dimensions—Aesthetics, Safety, Vibrancy, Wealth, Boring, and Depression—from street-view images;(2) used point-of-interest (POI) data from Dianping to classify street-level commercial activities into four major business types—Catering Services, Retail and Trade Locations, In-store Service Sectors, and Leisure and Experience Venues—and employed consumer ratings as a proxy indicator of satisfaction; and(3) adopted the Random Forest (RF) model in combination with SHapley Additive exPlanations (SHAP) to quantify the relative importance and influence effects of emotional perception variables across different business types. At the theoretical level, the study integrates insights from servicescape theory (Bitner, 1992 ), public space vitality theory, and emotional geography (Mehta, 2013 ) to conceptualize the pathway from physical-space emotional perception to psychological cognition. At the practical level, by revealing the differentiated emotional drivers of satisfaction across business categories, this research provides empirical evidence for the design of emotionally responsive pedestrian environments in mixed-use streets, as well as actionable decision-support tools for street design and urban spatial governance. 2 Literature review 2.1Reframing the Streetscape as an Open-Air Service Scape Bitner's (1992) servicescape theory established the foundational S–O–R (Stimulus–Organism–Response) framework in environmental behavior research. It posits that the physical environment shapes individuals' cognitive and affective responses through atmospheric, spatial, and symbolic cues, thereby influencing behavioral outcomes such as satisfaction, approach, and avoidance. As a core theory in service marketing, traditional research has primarily focused on controllable indoor commercial environments—such as stores and restaurants—where managers are assumed to systematically manipulate environmental elements to optimize customer experience. In recent years, an increasing number of interdisciplinary studies have extended servicescape theory to the analysis of street environments and street-level commerce. This body of research demonstrates the applicability of the atmospheric concept to outdoor streetscapes: the adjacent streetscape constitutes an open-air service interface for street commerce (Whyte, 1980 ) Its architectural façades, greenery, and pedestrian activities collectively form an open atmospheric system that reinforces the mutual visibility of "seeing and being seen." These spatial and social cues act on consumers' emotional perception before, during, and after the service encounter, thereby affecting overall satisfaction (Chen et al., 2020 ; Verhoef et al., 2009 ) this expanded paradigm, the street environment is no longer a neutral backdrop but an active service space that stimulates users' agency through multisensory stimuli, informal interactions, and public visibility (Mehta, 2013 ). The permeability between indoor and outdoor boundaries challenges the conventional management-control assumptions of servicescape theory and calls for an extended understanding of "distributed co-production" within the service atmosphere (Gehl, 2011 ). Consequently, the modern streetscape has evolved into a hybrid socio-spatial system in which environmental design, behavioral expression, and emotional perception co-construct the quality of urban experience. This theoretical shift challenges the core premises of the original servicescape framework—namely, the assumptions of clear environmental boundaries, unitary subjects, and centralized control—and highlights the co-constructed and socially embedded nature of street atmospheres. Reconstructing an expanded framework capable of explaining emotional experience in outdoor, dynamic, and co-governed environments has thus become a central research focus (Whyte, 1980 ). 2.2 Emotional Perception Dimensions and Commercial Satisfaction Within the servicescape theoretical framework, the physical environment primarily influences satisfaction and behavioral feedback through emotional perception. At the street scale, Lynch's (1960) concept of imageability reveals how street elements evoke collective atmospheres and shared memories((Lynch, 1960 ), while Jacobs's (1961) metaphor of the "sidewalk ballet" underscores the role of everyday interaction in fostering community safety and identity(Jacobs, 1961 ). The development of emotional geography further situates emotional experience within the relational and cultural fabric of place, conceptualizing the street as a dialectical unity of material and social processes (Mehta, 2013 ).Drawing on the Mehrabian–Russell (1974) model, prior research has established an emotional–behavioral chain of "environmental stimuli → affective state (e.g., safety, pleasure) → satisfaction and approach behavior," bridging micro- (individual), meso- (street), and macro-level (urban) studies on spatial vitality (Foroudi et al., 2020 ; Shukri et al., 2022 ) (Yi & Gim, 2018 ). This paradigm shift from physical determinism to emotional mediation marks the transition of street design research on spatial vitality from correlation toward explanation. However, existing studies often exhibit a form of typological blindness—treating "commercial satisfaction" as a homogeneous construct and conducting analyses aggregated at the street or district level (Walsh et al., 2011 ). Recent empirical findings indicate that perceived safety and walkable streetscape attributes associated with affective responses such as pleasure and aesthetics play a crucial role in shaping satisfaction within catering contexts (Koo et al., 2023 ) (Han et al., 2025 ). Nevertheless, few studies have examined the streetscape as a servicescape influencing differentiated emotional pathways across distinct business types, including catering, retail, cultural, and entertainment sectors (Ho et al., 2021 ; Stroebele & De Castro, 2004 ) example, catering services are highly dependent on warmth and social comfort (Heung & Gu, 2012); retail environments rely on perceived order and safety (Correia & Kozak, 2016 ༉; while cultural and leisure spaces emphasize aesthetic pleasure and place identity (Domínguez-Quintero et al., 2021 ). This differentiation highlights the need for streetscape emotion–commercial satisfaction models to recognize the heterogeneity of business functions and to explore contextual variations in emotional influence mechanisms. Such inquiry contributes both theoretical and empirical foundations for refined placemaking and governance in multifunctional street environments. 2.3 Multi-Source Heterogeneous Data Analysis Supporting Large-Scale Urban Perception and Cognition Research The rise of multi-source heterogeneous urban data has advanced large-scale research on urban perception and cognitive mechanisms. By integrating street-view imagery with point-of-interest (POI) data, scholars have been able to link objective dimensions of the built environment—such as green-view index, enclosure, and façade continuity—and functional attributes including business type, density, and diversity (Zhang et al., 2023 ) (Zhou et al., 2022 ), with subjective dimensions of emotional perception (e.g., safety, beauty, vibrancy) and commercial satisfaction indicators such as online ratings (Yang & Zhang, 2024 ) (Salesses, 2012 ). Deep learning models, particularly convolutional neural networks (CNNs), have enabled the automated extraction of emotional cues from visual data. The MIT Place Pulse project pioneered this approach by collecting global street-view images and crowdsourced perceptions of safety, beauty, and vibrancy (Naik et al., 2014 ). Subsequent studies have trained deep neural networks on these datasets to generate large-scale spatial emotion maps, revealing significant correlations between perceived safety and socio-economic indicators such as crime rates and housing prices, and enhancing the reliability of cross-cultural emotion prediction (van Veghel et al., 2024 ; Wei et al., 2022 ). Parallel to advances in visual analytics, diverse urban big data sources—including social media reviews (Ashkezari-Toussi et al., 2019 ; Despotovic & Hauser, 2025 ) (Zhou et al., 2018 ), pedestrian and location trajectories (Huang et al., 2018 ), mobile payment data (Feizizadeh et al., 2024 ), real-time pedestrian sensing (Cuesta-Mosquera et al., 2020 ; Qin et al., 2025 ), and thermal heat maps (Du et al., 2024 )—have been widely employed to measure collective emotional atmosphere, functional cognition, and behavioral trends. Together, these data construct the structural characteristics of emotion and cognition under a bidirectional spatial–behavioral mapping framework (Dubey, 2016; Luo et al., 2022 ), providing a quantitative foundation for examining the complex relationships between emotional perception and commercial satisfaction through large-scale data analysis. In terms of methodology, the Random Forest (RF) algorithm provides a robust tool for capturing nonlinear relationships and variable interactions often overlooked by traditional regression frameworks (Ma, 2023 ; Zhu et al., 2024 ). Unlike Geographically Weighted Regression (GWR), which assumes spatial linearity, RF accommodates complex hierarchical structures and cross-variable effects while minimizing overfitting. The SHapley Additive exPlanations (SHAP) approach further enhances interpretability by quantifying each variable's marginal contribution to model output (Chen et al., 2025 ; Luo et al., 2022 ). Collectively, these methodological advances facilitate the identification of dominant emotional predictors and their industry-specific variations, providing a powerful analytical pathway to decode emotion–satisfaction associations in mixed-use urban streetscapes. 3 Materials and Methods 3.1 Research Framework This study followed an integrated five-stage framework encompassing data construction, sample balancing, model training, model interpretation, and result analysis (see Fig. 1 ). In the data construction stage, the study selected the area enclosed by the Third Ring Road of Wuhan, Hubei Province, China, as the research boundary. Based on OpenStreetMap (OSM) street centerlines and Dianping point-of-interest (POI) data, POI types, ratings, and review counts were extracted within a 50 m buffer around each street centerline (Zhang et al., 2023 ). All POIs were categorized into four macro-level commercial sectors following the GB/T 4754–2017 and NAICS classification standards (Standardization Administration of & National Bureau of Statistics of, 2017), representing the primary business typologies along the streets. Simultaneously, the study combined the Place Pulse 2.0 dataset with Baidu Street View imagery ( https://lbsyun.baidu.com ). Using a pre-trained ResNet50 convolutional neural network, four-directional street-view images were processed to extract emotional perception scores across multiple affective dimensions at each sampling point. Each POI was then used as the center of a circular buffer with a 300 m radius—approximately corresponding to a 3–5 min walking distance (Koo et al., 2023 ). This spatial scale effectively covered the primary street segments in front of and behind catering venues and reflected the overall visual experience of visitors during their arrival and departure. Subsequently, all street-view images within each buffer were collected, and their perceptual features were averaged to represent the integrated emotional environment surrounding the corresponding POI (Han et al., 2025 ; Zhang et al., 2024 ) this process, a fine-grained, multi-source dataset integrating streetscape emotional perception and commercial satisfaction was constructed, providing a robust empirical foundation for subsequent model analyses. During the sample balancing stage, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to augment underrepresented categories, while random down-sampling was conducted on overrepresented ones. This ensured an equal number of samples across all categories, providing a balanced dataset for subsequent model training. In the model training stage, a Random Forest Regression (RF) model was employed, using six streetscape emotional perception indicators as independent variables and the commercial satisfaction rating as the dependent variable. Seventy-five percent of the samples were used for training and the remaining 25% for validation. To ensure robustness, the model underwent 100 iterations of random sampling and fitting. This approach allowed the exploration of nonlinear relationships and complex interaction effects embedded in the streetscape emotional data. During the model interpretation stage, the study adopted the Tree SHAP algorithm and utilized multiple visualization techniques, including ridge density plots, permutation importance boxplots, partial dependence plots (PDP), interaction heatmaps, and SHAP beeswarm plots. These visual representations facilitated a detailed analysis of how each perception indicator influenced commercial satisfaction. The figures illustrated the relative importance of features, nonlinear response patterns, inter-feature interactions, and marginal contributions to model output, thereby providing a transparent visualization framework for understanding the complex association between streetscape emotional perception and commercial satisfaction. The result analysis was conducted across four dimensions:(1) describing the distribution of commercial satisfaction across different business types;(2) assessing the relative importance of six perception indicators;(3) revealing nonlinear responses of perception indicators through PDP analyses; and(4) quantifying and visualizing feature interactions to elucidate the multifaceted effects of streetscape emotional perception on commercial satisfaction. 3.2 Study Data 3.2.1 Point of Interest (POI) Data In research on streetscape emotions, Points of Interest (POIs) serve as spatial representations of the "third place" beyond home (the first place) and work (the second place), encompassing named locations such as parks, cafés, and bookstores that accommodate social, leisure, and recreational activities (Psyllidis et al., 2022 ). With the advancement of digital technologies, POIs—expressed through geographic coordinates—have become proxy variables for real-world places and are widely applied in navigation, social activity analysis, and spatial research (Dhakal & Khadka, 2021 ; Liu et al., 2020 ). The rise of crowdsourced platforms such as Yelp, Foursquare, and OpenStreetMap (OSM) has further enriched POI diversity and attribute dimensions (e.g., operating hours, ratings, and photos), offering unprecedented spatial-semantic granularity for urban research (Gao et al., 2017 ). In China, Dianping has emerged as one of the leading local lifestyle service platforms, featuring an extensive user base and rich POI data. Since its establishment in 2003, Dianping has rapidly evolved into a key digital infrastructure for urban life, covering POI information across sectors such as dining, entertainment, tourism, and retail (Qin et al., 2019 ; Xu et al., 2018 ; Zhang et al., 2025 ). In this study, POI data were collected from Dianping within a 50 m buffer of street centerlines across the area enclosed by Wuhan's Third Ring Road(see Fig. 2 ). The dataset included 17 primary categories and 221 secondary subcategories. Among these, four business types were identified as having significant associations with streetscape perceptual environments; therefore, the analysis focused on experience-oriented business types. Catering Services (CAT) include restaurants, cafés, and other food-related venues. Customer satisfaction in this sector is closely linked to physical settings, ambiance, and service quality (Wang et al., 2023 ). This category was selected because its customer experience strongly depends on environmental cues and emotional engageme. Retailre visual components such as interior design and lighting directly shape emotional perception. Retail and Trade Locations (RTL) comprise supermarkets, home furnishing, and appliance stores—spaces frequently visited in daily life. Satisfaction within this category is jointly influenced by product attributes and visual atmosphere, consistent with retail environment theories emphasizing the direct i. Inct of aservice sectorserchandising on consumer behavior (Luo & Wang, 2003 ; McKenzie & Romm, 2021 ).In-store Service Sectors (ISS) include businesses such as beauty salons and healthcare services (Kleeman et al., 2023 ). Satisfaction in this category is primarily driven by accessibility and by the intensive interaction between customers and the physical setting during service delivery. T. Leisureunding streetscape exerts a notable influence on consumer decisions and experiential quality.Leisure and Experience Venues (LEX) encompass immersive entertainment spaces such as karaoke bars and family activity centers (Yang et al., 2020 ). Satisfaction in this category depends heavily on ambiance and social experience, with environmental elements like decoration and lighting playing a crucial role in shaping emotional engagement. Under the research objective of examining the influence mechanisms of external streetscape cues on adjacent commercial satisfaction, two types of POIs were excluded. First, function-oriented POIs (e.g., educational training centers, guesthouses, medical facilities) were removed because customer satisfaction in these sectors is primarily driven by professional quality and operational efficiency, with only marginal influence from streetscape characteristics. Second, nominal POIs that merely serve as place markers without any service-related experience were also excluded, as they have no substantive value for emotional perception modeling. After data screening and remapping, a total of 25,065 valid POI entries were retained. Following the GB/T 4754–2017 and NAICS classification standards, the dataset was categorized into four macro-level business types: Catering Services (CAT, 39.74%), Retail and Trade Locations (RTL, 4.05%), In-store Service Sectors (ISS, 41.58%), and Leisure and Experience Venues (LEX, 14.63%). Given the pronounced imbalance in sample sizes among the four categories, random undersampling was applied to reduce overrepresented classes (He & Garcia, 2009 ), while the Synthetic Minority Oversampling Technique (SMOTE) was employed to augment underrepresented ones (Chawla et al., 2002 ). The final dataset contained 3,000 samples per category, ensuring uniform sample size across business types. This classification framework focuses on environmentally sensitive, experience-oriented business sectors. Through rigorous selection and resampling, the dataset achieved structural balance, providing a consistent and reliable input for subsequent Random Forest and Tree-SHAP analyses. 3.2.2 Street-view images of Wuhan Street-view images of the study area were obtained from the Baidu Maps Static API ( https://lbsyun.baidu.com/ ). The platform has continuously archived street-view imagery since 2013, with many locations regularly updated. As of 2019, street-view samples were collected along the road network at 200 m intervals in four orientations—0, 90, 180, and 270°—yielding a total of 21,095 sampling points(see Fig. 3 ). Consistent with the Place Pulse 2.0 dataset, each street-view image was uniformly resized to 400 × 300 pixels to ensure standardized input dimensions for model processing. 3.2.3 The Place Pulse 2.0 dataset The Place Pulse 2.0 dataset, developed by the Massachusetts Institute of Technology (MIT), is a crowdsourced platform designed to collect human perceptual evaluations of urban environments (Naik et al., 2014 ). Through pairwise image comparisons conducted by online volunteers, the platform poses questions such as "Which place looks safer?" and aggregates large-scale perceptual data from street-view imagery. Compared with the first version, Place Pulse 2.0 substantially expands both the scope and depth of the dataset, encompassing 110,988 street-view images from 56 cities across 26 countries, and evaluates them along six perceptual dimensions: Safety, Aesthetics, Vibrancy, Wealth, Depression, and Boring. In this study, the Place Pulse 2.0 dataset provided the foundational data for the emotional perception model. Each street-view image was labeled according to its perceived score on these dimensions, calculated through pairwise comparison statistics. Specifically, the perceptual score of each image was derived from its relative win–loss frequency across multiple comparisons. These scores were further standardized using a Q-score normalization process, yielding a continuous scale ranging from 0 to 10((Wei et al., 2022 ). The dataset has played a significant role in various fields, particularly in urban planning (van Veghel et al., 2024 ), environmental psychology(Qiu et al., 2025 ), and street-view perception studies (Liu et al., 2023 ). Its extensive geographic coverage and multidimensional perceptual attributes make it a valuable resource for analyzing how urban environments influence human perception. By examining these perceptual data, researchers can better understand how specific visual characteristics of streetscapes shape public evaluations of safety, vibrancy, and related affective dimensions, thereby providing theoretical support for urban design and planning. 3.3 Data Processing and Method 3.3.1 Mapping human perception of urban landscape using ResNet50 In deep learning, the depth of a neural network—defined by the number of layers—is critical to its performance. Theoretically, deeper networks possess stronger representational capacity and can learn more complex features. However, in practical training, increasing network depth often leads to two major challenges:(1) Vanishing/Exploding Gradients: During backpropagation, gradients may exponentially decay or grow as the number of layers increases, making deep networks difficult to train.(2) Degradation Problem: Even when gradient issues are mitigated (e.g., through Batch Normalization), empirical results show that simply deepening the network may increase both training and testing errors, resulting in performance degradation. To address these issues, He et al. ( 2016 ) introduced skip connections, allowing the network to learn residual mappings rather than direct transformations.(He et al., 2016 ) Compared with conventional convolutional (Conv) or fully connected (FC) layers, the Residual Block design facilitates gradient propagation and improves optimization stability. For neural network–based regression, both model architecture and data quality are essential. As often stated, data define the upper limit of a model, while algorithms merely approximate it. Neural networks essentially learn statistical patterns from large datasets. In supervised learning, training data provide: (1) input–output correspondences that enable function approximation; (2) distributional information that helps the model understand diversity and latent structures; and (3) opportunities for feature extraction, allowing the model to automatically learn useful representations from variable combinations. In general, larger datasets tend to enhance model generalization, whereas low-quality data can significantly degrade performance. In supervised image regression tasks, labels must accurately reflect perceptual reality. Because visual qualities of photographs are difficult to describe textually, we used numerical scoring to enable neural representation, transforming image evaluation into a mathematical regression problem. However, label definition itself introduces subjectivity. To reduce bias, we manually annotated the data through pairwise image comparisons, assigning relative perceptual judgments that were later converted into numerical scores via algorithmic calibration. We thus reconstructed human perception representation as an image-based learning problem. First, street-view images with verified perceptual labels were converted into paired datasets with relative scores to build a standardized training set. Second, deep learning models were trained to extract features from labeled images. Finally, the trained model was applied to predict perceptual scores for new images within the study area. The resulting scores quantitatively represented human perception. Because correlations among perceptual dimensions were weak, each dimension was trained independently using six separate models. We conducted comparative experiments using AlexNet, VGG, and ResNet architectures. Results indicated that ResNet, with residual connections, significantly outperformed conventional CNNs of comparable depth in regression accuracy (Krizhevsky et al., 2012 ; Simonyan & Zisserman, 2015 ). Accordingly, ResNet was selected for this regression task. Among ResNet18, ResNet50, and ResNet101, experiments showed that ResNet18 lacked sufficient representational capacity due to limited parameters, while ResNet101 required excessive computational resources and was prone to instability and overfitting with our dataset. Thus, ResNet50 was adopted as the optimal architecture. Seventy percent of the dataset was used for training, with 60% as the training subset and 10% as the validation subset. The remaining 30% was reserved for model evaluation. The network was initialized with pretrained ResNet50 weights and fine-tuned for regression. The pretrained model was fine-tuned with an initial learning rate (lr) set to 1e − 6, which was dynamically adjusted during training. The original ResNet output layer, designed for classification, was retained, and a fully connected (FC) regression layer was appended to predict continuous values. The cross-entropy loss function was used as the training objective to stabilize gradient descent, while model validation and testing performance were assessed using the mean relative error between predicted and true labels. Training was performed on an NVIDIA RTX 2080Ti GPU with a batch size (batch_size) of 32 and a maximum of 100 epochs. Early stopping was applied when validation performance deteriorated or plateaued to prevent overfitting and improve generalization stability. 3.3.2 Random Forest regression Random Forest (RF) regression is an ensemble learning method that fits multiple decision tree regression models on randomly sampled subsets of both data and features, and then aggregates their outputs by averaging (Breiman, 2001). Prior studies have demonstrated that RF regression can effectively capture nonlinear relationships among variables (Gao et al., 2023 ; Wang et al., 2022 ). Therefore, this study also employed RF regression to examine the relationships between street-view emotional perception dimensions and commercial satisfaction, including the relative importance of variables and their partial dependence relationships. In this study, 75% of the samples were used for model training and the remaining 25% for validation. To ensure robustness, the random sampling and model fitting procedures were repeated 100 times, and the averaged outcomes were used for analysis. The coefficient of determination (R²) and the root mean square error (RMSE) were adopted as performance evaluation metrics for the model. 3.3.3 SHapley Additive exPlanations Although Random Forest (RF) regression can effectively reveal nonlinear relationships among variables, it has certain limitations in capturing interaction effects, which are crucial in understanding the determinants of commercial satisfaction. The SHAP algorithm originates from the Shapley value concept in cooperative game theory, which aims to fairly distribute payoffs among participants in a coalition. In machine learning, SHAP assigns each feature a contribution score (i.e., SHAP value) that quantifies its influence on model outputs (Lundberg & Lee, 2017 ) SHAP, a variant of the SHAP algorithm, is particularly well-suited for tree-based models such as Random Forests. It further decomposes feature contributions into main effects and interaction effects (Lundberg et al., 2018; Ma, 2023 ). Therefore, this study employed the Tree SHAP algorithm to detect the interactions among street-view emotional perception dimensions and their effects on commercial satisfaction, as well as to quantify the proportion of contribution attributable to feature interactions. 4 Results 4.1 Statistical Description of Commercial Satisfaction and Streetscape Perception Descriptive statistics of commercial satisfaction and streetscape perception were first computed for the four business types—Catering Services (CAT), Retail and Trade Locations (RTL), In-store Service Sectors (ISS), and Leisure and Experience Venues (LEX) (see Table 1 ). Results indicated that Aesthetics (Aes) showed relatively small mean variation across business types. Both CAT and RTL exhibited mean Aesthetics (Aes) scores above 4.85 with low standard deviations, suggesting consistent aesthetic evaluations among customers. For Safety (Saf), the mean score of CAT (3.06) was slightly higher than that of the other business types, indicating a more balanced perception of safety within catering environments. The mean Vibrancy (Vib) scores were similar across business types (around 3.4) but showed large standard deviations, suggesting substantial variation in perceived activity levels among respondents. Wealth (Wth) showed relatively high mean values (≈ 5.15) and small standard deviations, implying stable and consistent perceptions across business types. Depression (Dpr) also displayed high mean values (≈ 5.83) and low variability, reflecting a cross-type consistency in this perception. In contrast, Boring (Bro) had a lower mean of approximately 5.2 with larger standard deviations, indicating higher individual variability in the perception of boredom. Regarding commercial satisfaction, significant differences were observed among business types. The CAT sector showed the highest mean satisfaction (M = 3.74), exceeding that of the RTL sector (M = 3.49). RTL also exhibited a larger standard deviation, suggesting greater heterogeneity in satisfaction across customers within this category. As illustrated in Fig. 4 , the satisfaction distribution of CAT was more concentrated, with a sharp peak, indicating generally high and consistent satisfaction among customers. The ISS category displayed a flatter curve with a lower peak, suggesting a more dispersed evaluation pattern without a clear concentration. The LEX category showed a bimodal distribution, implying the presence of two distinct customer groups with divergent satisfaction levels. By contrast, the RTL category exhibited a wide and shallow distribution, suggesting both lower overall satisfaction and higher intra-group variability. Overall, the results demonstrate observable fluctuations in both streetscape perception and commercial satisfaction across business types, particularly in dimensions such as aesthetics, safety, and boredom. These findings highlight the subjective and diverse nature of customer perception and suggest that street design and business configuration should adopt differentiated optimization strategies tailored to the perceptual characteristics of each commercial type. Table 1 Statistical Description Across Different Business Models. Macro / Metric CAT ISS LEX RTL Aes 4.858 ± 0.066 4.850 ± 0.067 4.842 ± 0.068 4.858 ± 0.068 Saf 3.065 ± 0.089 3.055 ± 0.092 3.053 ± 0.090 3.064 ± 0.089 Vib 3.400 ± 0.120 3.410 ± 0.119 3.409 ± 0.120 3.406 ± 0.121 Wth 5.155 ± 0.086 5.150 ± 0.089 5.149 ± 0.089 5.143 ± 0.084 Dpr 5.837 ± 0.105 5.836 ± 0.109 5.830 ± 0.106 5.833 ± 0.105 Bro 5.227 ± 0.116 5.218 ± 0.122 5.217 ± 0.127 5.213 ± 0.117 Star rating 3.739 ± 0.347 3.717 ± 0.457 4.016 ± 0.462 3.490 ± 0.444 4.2 Relative importance analysis To enhance the robustness and reproducibility of the findings, beyond the relative importance ranking illustrated in Fig. 5 , this study further examined the contribution patterns of each emotional perception dimension across different business types using SHAP distribution analysis. Overall, the density peaks (density_peak) of all four business categories slightly fell on the left side of the zero axis (approximately − 0.05 to − 0.01), with all skewness values being positive. This distributional pattern indicates a shared tendency characterized by a near-zero or mildly negative baseline contribution coupled with a distinctly elongated positive tail. The combination of a "slightly negative peak with positive skewness" suggests that emotional dimensions generally exert mild net effects on satisfaction for most samples but exhibit asymmetric positive surges in certain contexts, thereby amplifying their overall explanatory power. In Catering Services (CAT), Depression (Dpr) and Vibrancy (Vib) demonstrated the strongest individual heterogeneity, with coefficients of variation (CV) of 156.13 and 183.46, respectively, both markedly higher than those of other dimensions. Their interquartile ranges (IQR) were also among the largest (Dpr = 0.046; Vib = 0.045). In contrast, Aesthetics (Aes) and Safety (Saf) exhibited much lower CVs (30.56 and 58.73) and narrower IQRs (0.039 and 0.043), implying greater stability in their contribution directions. All five dimensions showed positive skewness (e.g., Aes and Saf ≈ 1.83), which, together with the mildly negative density peaks, formed a "moderate central tendency with right-tail enhancement" pattern. In In-store Service Sectors (ISS), Aesthetics displayed the most pronounced distributional fluctuation, with an IQR of 0.104, far exceeding other dimensions (Wealth = 0.080; Safety = 0.055; Vibrancy = 0.052; Depression = 0.052), and a skewness of 1.37, suggesting significant bidirectional expansion and right-tail amplification. Meanwhile, Aesthetics and Wealth had relatively low CVs (19.55 and 24.78), indicating a "highly volatile but internally consistent" pattern. In contrast, Vibrancy showed a high CV (94.36), suggesting that although its overall contribution remained close to the zero axis, considerable fluctuation occurred at specific points. The safety distribution was more stable (IQR = 0.055; skewness = 0.42), with values concentrated in the non-negative range. In Leisure and Experience Venues (LEX), Aesthetics again showed the greatest volatility (IQR = 0.089; skewness = 1.59), followed by Safety and Depression (IQR = 0.078 and 0.075, respectively). Vibrancy had the smallest IQR (0.068) but a relatively high CV (69.51), indicating that while its overall effect hovered near zero, a certain degree of heterogeneity persisted. The density peaks of all dimensions remained mildly negative (approximately − 0.026 to − 0.019), which, combined with positive skewness, yielded a "broadly bidirectional but right-tail dominant" distributional pattern. In Retail and Trade Locations (RTL), Aesthetics exhibited the highest volatility and heterogeneity across all samples, with an IQR of 0.091, a CV of 266.91, and a skewness of 2.44, revealing pronounced right-tail amplification and inter-sample variability. Vibrancy also showed wide distributional spread (IQR = 0.077; skewness = 2.31). Although Safety and Wealth exhibited relatively narrow IQRs (0.054 and 0.052), their extreme right-tail skewness values (5.17 and 5.49, respectively) suggested the presence of rare but exceptionally strong positive enhancements. Depression occupied a moderate position (IQR = 0.069; CV = 104.73; skewness = 3.04). 4.3 Dependency between Driving Factors and Commercial Satisfaction Partial dependence plots (PDPs) revealed nonlinear relationships between each emotional perception dimension and commercial satisfaction (see Fig. 6 ). Results are reported by business type. In the catering services sector, Aesthetics (Aes) exhibited a nonlinear positive relationship with commercial satisfaction. As Aes approached 4.8, satisfaction began to increase gradually, and a sharp rise occurred between approximately 4.9 and 5.0, indicating a clear threshold effect in the influence of aesthetics on satisfaction. Vibrancy (Vib) showed a positive relationship with satisfaction; satisfaction increased steadily from 3.2 to 3.4, and the rate of increase accelerated near 3.4, indicating a threshold-like acceleration. Wealth (Wth) displayed nonlinear variation: satisfaction decreased between 5.0 and 5.2, yet rose sharply as Wth approached 5.2, evidencing a threshold effect. Depression (Dpr) also varied nonlinearly, with pronounced fluctuations in the 5.6–5.8 interval. A value of 5.7 was identified as a critical threshold: beyond this point satisfaction declined rapidly and became more volatile, after which the downward trend moderated but remained at a lower level. In the in-store service sectors sector, Aesthetics (Aes) exhibited marked nonlinear volatility. As Aes approached 4.8–4.9, satisfaction increased sharply and then entered a period of pronounced fluctuation; in particular, satisfaction dropped notably near 4.9 before rebounding, identifying 4.9 as a key threshold. Safety (Saf) showed a comparatively stable relationship with satisfaction: within 3.0–3.2 satisfaction varied little and no clear threshold effect was observed. Vibrancy (Vib) displayed a positive, nonlinear relationship: satisfaction rose steadily from 3.2 to 3.4, with a notable acceleration at 3.4, indicating 3.4 as a potential threshold where increases in Vib produce accelerated gains in satisfaction. Wealth (Wth) showed nonlinear change: from 5.0 to 5.2 satisfaction rose gradually, yet the rate increased sharply near 5.2, suggesting 5.2 as a critical threshold. Depression (Dpr) exhibited pronounced nonlinear variation in the 5.8–6.0 interval; satisfaction fell precipitously near 5.9–6.0, identifying 5.9–6.0 as critical thresholds. Boring (Bro) also varied nonlinearly: stability prevailed between 5.0 and 5.2, but significant volatility occurred approaching 5.2, and extreme variation appeared between 5.3 and 5.4. The 5.2–5.3 range was thus identified as a key threshold at which sustained, substantial changes in satisfaction occur. In the leisure and experience venues sector, Aesthetics (Aes) produced strong nonlinear effects: satisfaction was relatively stable from 4.8 to 4.9, but increased sharply from 4.9 to 5.0, with a pronounced jump near 4.9, indicating high sensitivity to aesthetic changes. Safety (Saf) showed limited fluctuation, primarily within 3.0–3.2, with small changes in satisfaction. Vibrancy (Vib) had only minor variation, concentrated in 3.2–3.4, implying a limited effect on satisfaction. Wealth (Wth) was positively associated with satisfaction: increases from 5.0 to 5.1 produced modest gains, whereas the rate of increase accelerated from 5.1 to 5.2, with 5.2 identified as a key threshold. Depression (Dpr) produced large satisfaction fluctuations in 5.8–6.0, with a sharp decline near 6.0; 5.9–6.0 were therefore threshold values. After an abrupt rise followed by a slight fall, the net effect was a weak positive association overall. Boring (Bro) was stable between 5.0 and 5.2, but showed significant volatility near 5.2 and a pronounced jump at 5.4; 5.2–5.3 was identified as the threshold range. In the retail and trade locations sector, Aesthetics (Aes) showed a marked negative association with satisfaction. While changes were modest near 4.8–4.9, satisfaction varied dramatically between 4.9 and 5.0, with a pronounced decline near 4.9, indicating 4.9 as a critical threshold beyond which fluctuations intensified. Safety (Saf) related positively to satisfaction: as Saf increased from 3.0 to 3.2, satisfaction gains accelerated, with 3.2 serving as a key threshold beyond which improvements in Saf produced faster satisfaction increases. Vibrancy (Vib) exhibited nonlinear volatility, particularly within 3.2–3.6, including multiple local peaks. Wealth (Wth) showed noticeable fluctuation in 5.0–5.2; near 5.1–5.2 satisfaction variation increased, and a rapid rise occurred approaching 5.2, although it did not return to the highest observed level. Depression (Dpr) displayed nonlinear fluctuation in 5.8–6.0, with a sharp decline near 6.0; 5.9–6.0 were identified as threshold values, and exceeding this range produced large increases in satisfaction variability. Boring (Bro) showed nonlinear changes, notably in 5.2–5.4; 5.3 emerged as a key threshold where satisfaction volatility began to increase, and near 5.4 the upward change in satisfaction moderated and ultimately declined. In summary, all four business categories exhibited distinct perceptual threshold ranges, among which the threshold effects of Aesthetics, Safety, and Wealth were the most pronounced. This finding suggests that emotional cues in streetscapes shape differentiated pathways influencing satisfaction across various commercial types. 4.4 Interaction Effects of Driving Factors on Commercial Satisfaction Based on the SHAP interaction analysis, the interactional contributions among emotional dimensions exhibited notable variations across different business categories. In the Catering Services (CAT) sector (see Fig. 7 a), Boring (Bro) demonstrated the strongest interaction intensity with other perceptual factors. The SHAP value approached 0.008, particularly in its interaction with Wealth (Wth) and Vibrancy (Vib), where commercial satisfaction fluctuated markedly. This indicates that social engagement exerts the most salient influence among all interactions. Strong interaction effects were also observed between Safety (Saf) and Depression (Dpr), Dpr and Vib, as well as wealth and Aesthetics (Aes), with SHAP values ranging from approximately 0.0075 to 0.008, suggesting that these interactions play an important role in explaining variations in satisfaction. In the In-store Service Sectors (ISS) category (see Fig. 7 b), the interaction between Wealth (Wth) and Aesthetics (Aes) showed the greatest intensity, with a SHAP value of 0.014, indicating that their joint effect had the most substantial influence on satisfaction. Additionally, interactions between Aesthetics (Aes) and Depression (Dpr), as well as between Aesthetics (Aes) and Boring (Bro), also demonstrated strong effects, with SHAP values of approximately 0.012, further emphasizing the centrality of aesthetics among perceptual dimensions. In the Leisure and Experience Venues (LEX) sector (see Fig. 7 c), the interactions between Aesthetics (Aes) and Vibrancy (Vib), and between Aesthetics (Aes) and Depression (Dpr), exhibited relatively high impacts, both with SHAP values exceeding 0.011. This suggests that these pairings significantly affect commercial satisfaction. Meanwhile, the interactions between Depression (Dpr) and Safety (Saf), Safety (Saf) and Wealth (Wth), Aesthetics (Aes) and Safety (Saf), as well as Boring (Bro) and Safety (Saf), also displayed strong effects, with SHAP values around 0.010, indicating that these combinations meaningfully influence satisfaction levels. In the Retail and Trade Locations (RTL) category (see Fig. 7 d), the interactions between Aesthetics (Aes) and Vibrancy (Vib), and between Aesthetics (Aes) and Depression (Dpr), exhibited the strongest intensities, with SHAP values exceeding 0.016. These combinations therefore exerted the most significant effects on satisfaction. Moreover, the interactions between Depression (Dpr) and Vibrancy (Vib), Boring (Bro) and Aesthetics (Aes), and Boring (Bro) and Wealth (Wth) also showed relatively strong influences, with SHAP values around 0.014, suggesting that these pairings contributed meaningfully to the observed variation in commercial satisfaction. Overall, Aesthetics, Safety, and Vibrancy consistently functioned as key driving dimensions across all business categories. Their influence pathways were not isolated or linear but were jointly shaped through the synergistic interplay of multidimensional perceptual cues that together defined the overall experience of commercial satisfaction. 5 Discussion 5.1Emotion-Driven Patterns across Business Categories This study employed a Random Forest model combined with SHAP interpretability analysis to systematically reveal, for the first time, the differentiated response mechanisms of multidimensional streetscape emotional perceptions on commercial satisfaction across distinct business categories, thereby overcoming the current research limitations of single-type and single-dimension analyses (Walsh et al., 2011 ). The Catering Services (CAT) sector demonstrated a distinctive "Depression-dominated and Vibrancy-synergized" pattern. Depression (Dpr) emerged as the most critical predictor, exhibiting a clear threshold effect on satisfaction: when Dpr scores exceeded 5.7, satisfaction declined sharply. This aligns with previous evidence that environmental stress significantly impairs dining experiences (Stroebele & De Castro, 2004 ), while the present study quantitatively identifies this perceptual turning point. Notably, Vibrancy (Vib), though the second most important predictor, showed a robust positive influence, particularly when Vib > 3.4, producing a marked enhancement effect. This finding supports Whyte's (1980) "outdoor interface" theory(Whyte, 1980 ༉, suggesting that catering environments rely on street vitality to create "watchability" yet must balance stimulation with pressure control to maintain comfort. The In-store Service (ISS) and Leisure and Experience Venues (LEX) sectors shared a "Aesthetics-centered driving" mechanism. In both categories, Aesthetics (Aes) ranked highest in SHAP importance, and its impact displayed pronounced bidirectionality: high Aes scores (Aes > 4.9) strongly elevated satisfaction, whereas low aesthetic values triggered sharp declines. This supports environmental psychology perspectives, which posit that in high-contact service settings such as salons and cafés, visual aesthetics directly shape consumer identity and self-congruity (Japutra et al., 2019 ). However, a key difference emerged: in the LEX sector, Depression (Dpr) ranked fourth in importance—significantly higher than in the ISS sector, where it was the least influential. The partial dependence plot (PDP) indicated that when Dpr > 5.9, satisfaction dropped rapidly, suggesting that immersive leisure environments are more sensitive to environmental stress, whereas functional service contexts show greater tolerance. The Retail and Trade Locations (RTL) sector exhibited an "aesthetic paradox." Although Aesthetics (Aes) ranked as the most important variable, its SHAP value distribution showed a pronounced negative skew, and the PDP revealed a clear satisfaction decline when Aes > 4.9. This finding runs counter to conventional assumptions but is consistent with visual complexity theory: moderate aesthetic richness enhances attention and pleasure, whereas excessive aesthetic or visual complexity may surpass a cognitive threshold, increasing cognitive load, reducing processing fluency, and elevating mismatched expectations, ultimately decreasing satisfaction (Reber et al., 2004 ). Empirical research on retail spaces likewise shows that high visual complexity often reduces pleasure among low-engagement consumers and, through emotional mediation, diminishes store entry and purchase intentions. Hence, the intuitive notion that "more aesthetics equals better experience" does not necessarily hold true and may exhibit threshold effects (Jang et al., 2018 ). Moreover, this study observed a notable satisfaction surge when Vibrancy (Vib) exceeded 3.4, suggesting that retail experiences rely more heavily on environmental dynamism and sensory stimulation to evoke immediate emotions and impulsive purchasing behaviors (Huang et al., 2018 ). Therefore, in RTL contexts, enhancing static aesthetics without integrating dynamic vitality may inadvertently suppress overall satisfaction. 5.2Business-Type Differentiation Mechanisms of Interaction Effects Interaction analysis further revealed the cross-dimensional emotional synergy mechanisms within each business category. In the Catering Services (CAT) sector, the interaction between Boring (Bro) and Wealth (Wth) was the strongest (SHAP = 0.008). High-wealth environments, such as premium dining districts, appeared to mitigate the negative impact of boredom through symbolic consumption. When Wth exceeded 5.2, visual markers such as luxury storefronts and branded façades functioned as "anchors of identity recognition," effectively reducing customers' sense of boredom. Conversely, in lower-wealth contexts such as mass food streets, a "density reinforcement" effect likely emerged, creating a lively and vibrant atmosphere ("renao") that offset monotony. This outcome may relate to social interaction patterns embedded in street activities; however, as cultural orientation was not directly measured in this study, cross-cultural data are required for further validation. In the In-store Services (ISS) sector, a strong interaction between Aesthetics (Aes) and Wealth (Wth) (SHAP = 0.014) was identified, reflecting a mechanism of signaling and credibility construction. The combination of high aesthetic design and refined facilities reduced uncertainty about service quality through dual "aesthetic–capital" cues, particularly in functional service venues such as hair salons and repair shops. However, under lower Wth conditions, excessive aesthetic design appeared to induce cognitive dissonance, suggesting that inconsistency between exterior design and perceived prosperity may trigger psychological conflict. Given that consumers typically spend short durations in such venues and rely heavily on visual information for rapid quality assessments, the synergy between aesthetics and wealth likely generates a multiplier effect in visual evaluation. In the Leisure and Experience Venues (LEX) sector, nonlinear interactions between Aesthetics (Aes) and Vibrancy (Vib) (SHAP = 0.011) were found to determine the degree of immersive experience. In highly aesthetic environments, customer engagement with visual stimuli intensified; however, when Vib surpassed a certain threshold, external stimulation began to compete for attentional resources, thereby undermining immersion. This finding aligns with the attentional resource competition model (Mehrabian & Russell, 1974 ), suggesting that overstimulation may counteract the intended experiential benefits of aesthetic investment. In the Retail and Trade Locations (RTL) sector, a significant negative interaction between Aesthetics (Aes) and Depression (Dpr) was observed (SHAP = − 0.016). In high-depression environments, high aesthetic levels amplified the sense of spatial oppression through a "cognitive contrast enhancement" effect. This was particularly evident in luxury malls, where refined decorations and crowded traffic jointly evoked consumer anxiety, supporting the notion of an "aesthetic oppression paradox." Conversely, moderate visual clutter in traditional markets appeared to reduce sensitivity to oppressive cues, suggesting a cultural tolerance for mild disorder in such contexts. These findings highlight the importance of balancing attractiveness and comfort in retail design: excessive aesthetic refinement may provoke negative emotional responses and consequently diminish overall commercial satisfaction. 5.3Theoretical Reconstruction and Contextual Innovation Grounded in servicescape theory and emotional geography, this study extends existing theoretical frameworks in three primary ways. First, it empirically validates an "open-air extension" of servicescape theory. Although the original model primarily emphasized environmental control within enclosed service settings, the present findings indicate that outdoor streetscapes possess distinctive emotional dynamics. While Safety (Saf) significantly affects satisfaction, socially oriented dimensions—particularly Vibrancy (Vib) and Boring (Bro)—exert even greater influence. This pattern supports the "street as open-air theater" concept (Mehta, 2013 ), in which informal social interactions (e.g., street vending, casual neighbor exchanges) constitute core mechanisms of affective generation. The results therefore suggest that managers of open urban environments should prioritize fostering harmonious social interactions rather than focusing solely on the control of physical spatial design. Second, the study empirically confirms the "emotion → satisfaction" transmission pathway proposed in emotional geography and simultaneously uncovers sectoral heterogeneity (Yüksel et al., 2010 ). Specifically, in the Catering Services (CAT) sector, emotional dimensions such as Depression (Dpr) and Vibrancy (Vib) significantly influenced customer satisfaction. In contrast, in the Retail and Trade Locations (RTL) sector, Aesthetics (Aes) emerged as the dominant determinant. These findings imply that different business types may exhibit distinct moderating mechanisms within their emotional influence pathways, warranting further investigation through multi-group or moderated structural equation modeling (SEM). Third, the results challenge two prevalent theoretical simplifications: (1) The "vitality supremacy" assumption. While Ewing et al. ( 2006 ) emphasized Vibrancy (Vib) as a core determinant of environmental appeal, the current findings reveal that its importance is highly contingent on business type((Ewing et al., 2006 ). In the Catering Services (CAT) sector, Vib ranked second in predictive importance, whereas in the Leisure and Experience Venues (LEX) sector, it ranked last and exhibited an upper threshold beyond which satisfaction declined. This suggests that urban design should aim for a balanced emotional composition rather than a singular pursuit of vitality. (2) The linearity assumption. Traditional OLS-based models have often posited a simple positive linear relationship between safety (Saf) and satisfaction (Jani & Han, 2011 ). However, the Partial Dependence Plots (PDPs) in this study reveal a nonlinear association between the two. For instance, in the Catering Services (CAT) sector, when Saf scores were below 3.2, their impact on satisfaction was negligible, but once Saf exceeded this threshold, the marginal effect increased sharply. A similar S-shaped pattern was also observed in the Retail and Trade Locations (RTL) sector, further demonstrating the generalizability of nonlinear relationships in affective perception research. 5.4 Emotion-Customized Spatial Intervention Strategies Building upon the results of the Random Forest model and SHAP interpretation, this study proposes an "Emotion-Customized" spatial intervention framework to support precision-based design strategies across different commercial environments. The findings demonstrate that the dominant emotional drivers vary significantly by business category. Therefore, urban design and commercial renovation should avoid applying uniform templates and instead adopt differentiated interventions according to the primary affective mechanisms of each sector. In the Catering Services (CAT) environment, model results indicated that Depression (Dpr) exhibited a clear inflection range between approximately 5.5 and 5.8, beyond which satisfaction declined sharply. This empirical threshold suggests the need for "stress-reducing" spatial interventions. Design strategies may focus on mitigating spatial enclosure and visual crowding by optimizing sight corridors, enhancing transparency of outdoor seating interfaces, and introducing localized greenery and natural lighting to foster a sense of visible relaxation and openness. In the In-store Services (ISS) and Leisure and Experience Venues (LEX) sectors, Aesthetics (Aes) and Wealth (Wth) displayed a synergistic relationship. When both dimensions reached higher levels, satisfaction improved markedly; however, excessive aesthetic refinement within low-wealth environments produced cognitive dissonance. Accordingly, design approaches should emphasize credibility through calibrated precision of visual and material details while avoiding ornamental excess. In the LEX sector specifically, Vibrancy (Vib) exhibited a nonlinear boundary within the 3.2–3.6 range, suggesting the necessity of controlling the upper limit of external stimulation. Techniques such as zoned acoustic and lighting modulation or buffered circulation pathways may balance immersive engagement with ambient vitality. In the Retail and Trade Locations (RTL) environment, the model identified a negative interaction between high Aesthetics (Aes) and high Depression (Dpr), where satisfaction declined substantially when both values were elevated. This pattern reveals a phenomenon of "aesthetic oppression". Consequently, design priorities should shift from aesthetic intensification toward behavioral comfort enhancement, restoring a sense of everyday liveliness through optimized façade openness, rhythm-controlled pedestrian flow, and tolerance for moderate visual irregularity within marketplace layouts. Overall, emotion-customized design underscores contextual differentiation rather than aesthetic uniformity. By translating data-driven emotional thresholds into actionable design parameters, this research bridges machine learning analytics with urban design practice, providing empirical grounding and theoretical advancement for data-informed affective urbanism. 5.5 Limitations and prospects This study has several methodological limitations.(1) Static imagery bias: The analysis relied on static Baidu street-view images collected in 2019, which excluded temporal variations. For instance, nighttime economic periods (18:00–24:00) with changing lighting conditions and vendor density, as well as seasonal transformations such as cherry blossom or rainy seasons, may substantially alter streetscape semantics but were not incorporated. To address this limitation, future research may integrate satellite-based nighttime light data with nocturnal street-view imagery to construct a spatiotemporal cube model, enabling the capture of dynamic changes in urban streetscapes.(2) Limited sensory modality: The study focused solely on the visual dimension, overlooking auditory factors (e.g., street vendor calls, traffic noise) and olfactory cues (e.g., food aroma, pollution) that likely affect emotional perception. Future efforts may deploy portable sensor networks to record multisensory data and combine these with social media sentiment analysis to build a multimodal emotional mapping framework, thereby achieving a more comprehensive understanding of how streetscapes shape emotional responses.(3) Black-box limitation of nonlinear mechanisms: Although the Random Forest model effectively captured threshold effects, it did not decompose the psychological pathways through which emotional perception influences satisfaction—such as place attachment or consumption-related affect. Subsequent studies could integrate Bayesian Networks (BN) with Structural Equation Modeling (SEM) to quantify the transmission weights along the emotion–function cognition pathway and further elucidate the complexity of underlying nonlinear mechanisms.(4) Cross-cultural adaptability of perceptual models: Cultural perception bias may influence ratings of aesthetics and safety. Therefore, model fine-tuning tailored to local linguistic and semantic contexts is necessary to ensure contextual validity.(5) User-generated content bias: Review data from the Dianping platform may be subject to selective self-reporting bias. Future research could integrate check-in or transaction data to enhance reliability and validity. Collectively, these methodological limitations provide valuable directions for future refinement and may contribute to a more precise understanding of the multidimensional mechanisms underlying emotional perception of urban streetscapes. 6 Conclusions This study extends the theory of service scape to open street environments by integrating street-view imagery, POI data, and machine learning methods. It systematically reveals the multidimensional mechanisms by which emotional perception of streetscapes influences commercial satisfaction and demonstrates differentiated patterns across business types. First, the findings identify business-type-specific emotional driving patterns. The Catering Services (CAT) sector exhibits a "Depression-dominated and Vibrancy-synergistic" mode, in which Depression (Dpr) serves as the primary negative predictor with a distinct threshold effect—once exceeding a critical value, satisfaction declines sharply. In contrast, Vibrancy (Vib) exerts a positive synergistic influence, particularly when surpassing a moderate level, leading to a marked increase in satisfaction. Both the In-store Service Sectors (ISS) and Leisure and Experience Venues (LEX) share an "Aesthetics-centered driving" mechanism, wherein Aesthetics (Aes) exerts a bidirectional impact on satisfaction. Higher aesthetic ratings significantly enhance satisfaction, whereas lower ratings cause a rapid decline. Notably, in the LEX category, Depression (Dpr) demonstrates a stronger negative impact on satisfaction than in ISS. The Retail and Trade Locations (RTL) sector presents an "aesthetic paradox," where excessive aesthetics (Aes) may instead reduce satisfaction. Conversely, improvements in Vibrancy (Vib) substantially enhance satisfaction, underscoring the reliance of retail environments on stimulus-driven consumption. Second, the study clarifies the business-type heterogeneity of interaction effects among emotional dimensions. The interrelationships among emotional variables are both significant and divergent across categories. In CAT, Boring (Bro) and Wealth (Wth) exhibit strong interactive effects that modulate customer satisfaction. In ISS, the interaction between Aesthetics (Aes) and Wealth (Wth) enhances perceived service credibility through dual "aesthetic–capital" cues. In LEX, the nonlinear interaction between Aesthetics (Aes) and Vibrancy (Vib) determines the maintenance of immersive experience. In RTL, a strong negative interaction between Aesthetics (Aes) and Depression (Dpr) reveals a "cognitive contrast amplification" effect, in which high aesthetic refinement under high-pressure conditions intensifies perceived spatial oppression. Theoretically, this study advances the conceptual boundaries of the service atmosphere theory and emotional geography in three ways. (1) The outdoor extension of service atmosphere theory. Classical models emphasize the controllability of enclosed commercial spaces, whereas this study demonstrates that the streetscape itself functions as an effective service scape. Satisfaction is influenced not only by safety (Saf) perception but also by social and informal interactional cues, validating the notion that nonformal interactions (e.g., vendors, street conversations) contribute substantially to emotional generation. (2) The operational verification of emotional geography across business contexts. The study quantitatively confirms that the "environmental emotion → place satisfaction" pathway varies across business types, indicating that the mechanism of emotional perception is jointly moderated by spatial functionality and consumption attributes. (3) The correction of linear assumptions and the introduction of nonlinear modeling. Using Random Forest and SHAP algorithms, the results reveal widespread threshold and nonlinear relationships between variables such as Safety (Saf), Aesthetics (Aes), and Vibrancy (Vib) and satisfaction, thereby overcoming the limitations of traditional linear models. In summary, supported by interpretable machine learning modeling, this study establishes a comprehensive analytical framework linking streetscape emotional perception to commercial satisfaction, unveiling the complex mechanisms through which urban streets generate affective and experiential responses. The findings provide data-driven and fine-grained decision-making references for street-space design and renewal, while offering practical and theoretical support for future exploration of "data-driven emotional design" and "business-type-customized spatial interventions." Declarations Funding This work was supported by the National Social Science Foundation Arts Project of China [Grant No. 22CG182] and the National Natural Science Foundation of China [Grant No. 52208088]. Author Contribution Yiwei Mo was responsible for the conceptualization and overall design of the study, performed data curation and management, conducted formal analysis, and drafted the original manuscript. Chao Luo and Jingjing Wang contributed to the writing of the original draft, was responsible for methodology design and data management, and secured project funding. Youpeng Lu contributed to the review and editing of the manuscript. Yunzhong Wang was in charge of project administration and coordination and participated in the review and editing of the manuscript. Qi Wu was responsible for visualization and assisted with formal analysis.Zixiang Feng was responsible for visualization and assisted with formal analysis. Data Availability Data will be made available on request. References Ashkezari-Toussi S, Kamel M, Sadoghi-Yazdi H (2019) Emotional maps based on social networks data to analyze cities emotional structure and measure their emotional similarity. Cities 86:113–124. https://doi.org/10.1016/j.cities.2018.09.009 Bitner MJ (1992) Servicescapes: The Impact of Physical Surroundings on Customers and Employees. J Mark 56(2):57–71. https://doi.org/10.1177/002224299205600205 Chawla NV, Bowyer KW, Hall LO, Kegelmeyer WP (2002) SMOTE: synthetic minority over-sampling technique. J Artif Intell Res 16:321–357 Chen S, Yu B, Shi G, Cai Y, Wang Y, He P (2025) Scale-Dependent Relationships Between Urban Morphology and Noise Perception: A Multi-Scale Spatiotemporal Analysis in New York City. Land 14(3). https://doi.org/10.3390/land14030476 Chen T-L, Chiu H-W, Lin Y-F (2020) How do East and Southeast Asian Cities Differ from Western Cities? A Systematic Review of the Urban Form Characteristics. Sustainability 12(6). https://doi.org/10.3390/su12062423 Correia A, Kozak M (2016) Tourists' shopping experiences at street markets: Cross-country research. Tour Manag 56:85–95. https://doi.org/10.1016/j.tourman.2016.03.026 Cuesta-Mosquera AP, Wahl M, Acosta-López JG, García-Reynoso JA, Aristizábal-Zuluaga BH (2020) Mixing layer height and slope wind oscillation: Factors that control ambient air SO2 in a tropical mountain city. Sustainable Cities Soc 52. https://doi.org/10.1016/j.scs.2019.101852 Despotovic M, Hauser C (2025) A beautiful place: investigating the determinants of perceived scenic beauty in Austrian landscapes with social media data. Humanit Social Sci Commun 12(1). https://doi.org/10.1057/s41599-024-04317-2 Dhakal CK, Khadka S (2021) Heterogeneities in Consumer Diet Quality and Health Outcomes of Consumers by Store Choice and Income. Nutrients 13(4). https://doi.org/10.3390/nu13041046 Domínguez-Quintero AM, González-Rodríguez MR, Roldán JL (2021) The role of authenticity, experience quality, emotions, and satisfaction in a cultural heritage destination. In Authenticity and Authentication of Heritage (pp. 103–117). https://doi.org/10.4324/9781003130253-9 Du F, Wang J, Mao L, Kang J (2024) Daily rhythm of urban space usage: insights from the nexus of urban functions and human mobility. Humanit Social Sci Commun 11(1). https://doi.org/10.1057/s41599-023-02577-y Dubey A, Naik N, Parikh D, Raskar R, Hidalgo CA (2016) Deep learning the city: Quantifying urban perception at a global scale. Lecture Notes in Computer Science Ewing R, Handy S, Brownson RC, Clemente O, Winston E (2006) Identifying and Measuring Urban Design Qualities Related to Walkability. J Phys Activity Health 3(S1):S223–S240. https://doi.org/10.1123/jpah.3.s1.s223 Feizizadeh B, Omarzadeh D, Blaschke T (2024) Spatiotemporal mapping of urban trade and shopping patterns: A geospatial big data approach. Int J Appl Earth Obs Geoinf 128. https://doi.org/10.1016/j.jag.2024.103764 Foroudi P, Cuomo MT, Foroudi MM, Katsikeas CS, Gupta S (2020) Linking identity and heritage with image and a reputation for competition. J Bus Res 113:317–325 Gao S, Janowicz K, Couclelis H (2017) Extracting urban functional regions from points of interest and human activities on location-based social networks. Trans GIS 21(3):446–467. https://doi.org/10.1111/tgis.12289 Gao Y, Zhao J, Han L (2023) Quantifying the nonlinear relationship between block morphology and the surrounding thermal environment using random forest method. Sustainable Cities Soc 91. https://doi.org/10.1016/j.scs.2023.104443 Gehl J (2011) Life between buildings Han C, Lieu SJ, Hwang U, Guhathakurta S (2025) Do streetscapes still matter for customer ratings of eating and drinking establishments in car-dependent cities? J Urban Des 1–22. https://doi.org/10.1080/13574809.2025.2541953 He H, Garcia EA (2009) Learning from imbalanced data. IEEE Trans Knowl Data Eng 21(9):1263–1284 He K, Zhang X, Ren S, Sun J (2016) Deep Residual Learning for Image Recognition Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Ho TP, Stevenson M, Thompson J, Nguyen TQ (2021) Evaluation of Urban Design Qualities across Five Urban Typologies in Hanoi. Urban Sci 5(4). https://doi.org/10.3390/urbansci5040076 Huang Z, Ling X, Wang P, Zhang F, Mao Y, Lin T, Wang F-Y (2018) Modeling real-time human mobility based on mobile phone and transportation data fusion. Transp Res Part C: Emerg Technol 96:251–269. https://doi.org/10.1016/j.trc.2018.09.016 Istrate A-L (2025) Street vitality: what predicts pedestrian flows and stationary activities on predominantly residential Chinese streets, at the mesoscale? J Plann Educ Res 45(1):66–80 Jacobs J (1961) The Death and Life of Great American Cities. Random House Jang JY, Baek EJ, Yoon SY, Choo HJ (2018) Store Design: Visual Complexity and Consumer Responses. Int J Des 12(2):105–118. https://www.ijdesign.org/index.php/IJDesign/article/view/2934 Jani D, Han H (2011) Investigating the key factors affecting behavioral intentions. Int J Contemp Hospitality Manage 23(7):1000–1018. https://doi.org/10.1108/09596111111167579 Japutra A, Ekinci Y, Simkin L (2019) Self-congruence, brand attachment and compulsive buying. J Bus Res 99:456–463 Kleeman A, Giles-Corti B, Gunn L, Hooper P, Foster S (2023) The impact of the design and quality of communal areas in apartment buildings on residents' neighbouring and loneliness. Cities 133. https://doi.org/10.1016/j.cities.2022.104126 Koo BW, Hwang U, Guhathakurta S (2023) Streetscapes as part of servicescapes: Can walkable streetscapes make local businesses more attractive? Comput Environ Urban Syst 106:102030 Krizhevsky A, Sutskever I, Hinton GE (2012) ImageNet Classification with Deep Convolutional Neural Networks. Advances in neural information processing systems Li X, Chen M, Wang R (2024) Assessing the nonlinear impact of green space exposure on psychological stress perception using machine learning and street view images. Front Public Health 12:1402536. https://doi.org/10.3389/fpubh.2024.1402536 Liu K, Yin L, Lu F, Mou N (2020) Visualizing and exploring POI configurations of urban regions on POI-type semantic space. Cities 99. https://doi.org/10.1016/j.cities.2020.102610 Liu Y, Chen M, Wang M, Huang J, Thomas F, Rahimi K, Mamouei M (2023) An interpretable machine learning framework for measuring urban perceptions from panoramic street view images. iScience 26(3):106132. https://doi.org/10.1016/j.isci.2023.106132 Lundberg SM, Lee S-I (2017) A unified approach to interpreting model predictions. Advances in neural information processing systems , 30 Luo P, Yu B, Li P, Liang P (2022) Spatially varying impacts of the built environment on physical activity from a human-scale view: Using street view data. Front Environ Sci 10. https://doi.org/10.3389/fenvs.2022.1021081 Luo W, Wang F (2003) Measures of Spatial Accessibility to Healthcare in a GIS Environment: Synthesis and a Case Study in Chicago Region. Environ Plann B Plann Des 30(6):865–884. https://doi.org/10.1068/b29120 Lynch K (1960) The Image of the City. MIT Press Ma Z (2023) Deep exploration of street view features for identifying urban vitality: A case study of Qingdao city. Int J Appl Earth Obs Geoinf 123. https://doi.org/10.1016/j.jag.2023.103476 McKenzie G, Romm D (2021) Measuring urban regional similarity through mobility signatures. Comput Environ Urban Syst 89. https://doi.org/10.1016/j.compenvurbsys.2021.101684 Mehrabian A, Russell JA (1974) An Approach to Environmental Psychology. MIT Press Mehta V (2013) The street: a quintessential social public space. Routledge Meng Q, Sun Y, Kang J (2017) Effect of temporary open-air markets on the sound environment and acoustic perception based on the crowd density characteristics. Sci Total Environ 601–602:1488–1495. https://doi.org/10.1016/j.scitotenv.2017.06.017 Mittal V, Kamakura WA (2009) The service quality–satisfaction link revisited: exploring asymmetries and dynamics. J Acad Mark Sci 37(3):378–390. https://doi.org/10.1007/s11747-009-0152-2 Naik N, Philipoom J, Raskar R, Hidalgo C (2014) Streetscore-predicting the perceived safety of one million streetscapes Proceedings of the IEEE conference on computer vision and pattern recognition workshops Psyllidis A, Gao S, Hu Y, Kim EK, McKenzie G, Purves R, Yuan M, Andris C (2022) Points of Interest (POI): a commentary on the state of the art, challenges, and prospects for the future. Comput Urban Sci 2(1):20. https://doi.org/10.1007/s43762-022-00047-w Qin L, Sun J, Niu Q, Yuan M (2025) Residential spatial differentiation and influencing factors of permanent and temporary populations based on mobile signaling data: A case study of Wuhan, China. Appl Geogr 184. https://doi.org/10.1016/j.apgeog.2025.103760 Qin X, Zhen F, Gong Y (2019) Combination of Big and Small Data: Empirical Study on the Distribution and Factors of Catering Space Popularity in Nanjing, China. J Urban Plan Dev 145(1). https://doi.org/10.1061/(asce)up.1943-5444.0000489 Qiu Y, Wu M, Huang Q, Kang Y (2025) Do You Know Your Neighborhood? Integrating Street View Images and Multi-task Learning for Fine-Grained Multi-Class Neighborhood Wealthiness Perception Prediction. Cities , 158 . https://doi.org/10.1016/j.cities.2025.105703 Reber R, Schwarz N, Winkielman P (2004) Processing fluency and aesthetic pleasure: Is beauty in the perceiver’s processing experience? Personality Social Psychol Rev 8(4):364–382. https://doi.org/10.1207/s15327957pspr0804_3 Salesses MP (2012) Place Pulse: Measuring the collaborative image of the city. Massachusetts Institute of Technology] Shukri SM, Wahab MH, Awaluddin ZL, Aminuddin AMR, Hasan MI (2022) The Role of Attachment in Creating Sustainable Sense of Place for Traditional Streets in Alor Setar, Malaysia. J Des Built Environ 22(1):55–71 Simonyan K, Zisserman A (2015) Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv preprint. Issue. https://arxiv.org/abs/1409.1556 Standardization Administration of, C., & National Bureau of Statistics of, C (2017) GB/T 4754–2017 国民经济行业分类 Stroebele N, De Castro JM (2004) Effect of ambience on food intake and food choice. Nutrition 20(9):821–838. https://doi.org/10.1016/j.nut.2004.05.012 van Veghel J, Dane G, Agugiaro G, Borgers A (2024) Human-centric computational urban design: optimizing high-density urban areas to enhance human subjective well-being. Comput Urban Sci 4(1). https://doi.org/10.1007/s43762-024-00124-2 Verhoef PC, Lemon KN, Parasuraman A, Roggeveen A, Tsiros M, Schlesinger LA (2009) Customer Experience Creation: Determinants, Dynamics and Management Strategies. J Retail 85(1):31–41. https://doi.org/10.1016/j.jretai.2008.11.001 Walsh G, Shiu E, Hassan LM, Michaelidou N, Beatty SE (2011) Emotions, store-environmental cues, store-choice criteria, and marketing outcomes. J Bus Res 64(7):737–744. https://doi.org/10.1016/j.jbusres.2010.07.008 Wang J, Gao C, Wang M, Zhang Y (2023) Identification of Urban Functional Areas and Urban Spatial Structure Analysis by Fusing Multi-Source Data Features: A Case Study of Zhengzhou, China. Sustainability 15(8). https://doi.org/10.3390/su15086505 Wang Y, Chen X, Gao M, Dong J (2022) The use of random forest to identify climate and human interference on vegetation coverage changes in southwest China. Ecol Ind 144. https://doi.org/10.1016/j.ecolind.2022.109463 Wei J, Yue W, Li M, Gao J (2022) Mapping human perception of urban landscape from street-view images: A deep-learning approach. Int J Appl Earth Obs Geoinf 112:102886 Whyte WH (1980) The Social Life of Small Urban Spaces. Conservation Foundation Xu F, Zhen F, Qin X, Wang X, Wang F (2018) From central place to central flow theory: an exploration of urban catering. Tourism Geographies 21(1):121–142. https://doi.org/10.1080/14616688.2018.1457076 Yang C, Zhang Y (2024) Public emotions and visual perception of the East Coast Park in Singapore: A deep learning method using social media data. Urban Forestry Urban Green 94. https://doi.org/10.1016/j.ufug.2024.128285 Yang R, Zhang J, Xu Q, Luo X (2020) Urban-rural spatial transformation process and influences from the perspective of land use: A case study of the Pearl River Delta Region. Habitat Int 104. https://doi.org/10.1016/j.habitatint.2020.102234 Yi Y-M, Gim T-H (2018) What Makes an Old Market Sustainable? An Empirical Analysis on the Economic and Leisure Performances of Traditional Retail Markets in Seoul. Sustainability 10(6). https://doi.org/10.3390/su10061779 Yüksel A, Yüksel F, Bilim Y (2010) Destination attachment: Effects on customer satisfaction and cognitive, affective and conative loyalty. Tour Manag 31(2):274–284. https://doi.org/10.1016/j.tourman.2009.03.007 Zhang D, Chen M, Huang W, Gong Y, Zhao K (2024) Exploring urban semantics: A multimodal model for POI semantic annotation with street view images and place names. Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence Zhang E, Hou J, Long Y (2025) The form of China’s urban commercial expansion in the digital era. Nat Cities 2(7):639–649. https://doi.org/10.1038/s44284-025-00254-6 Zhang E, Xie H, Long Y (2023) Decoding the association between urban streetscape skeletons and urban activities: Experiments in Beijing using Dazhong Dianping data. Trans Urban Data Sci Technol 2(1):3–18 Zhou H, Gu J, Liu Y, Wang X (2022) The impact of the skeleton and skin for the streetscape on the walking behavior in 3D vertical cities. Landsc Urban Plann 227:104543 Zhou X, Noulas A, Mascolo C, Zhao Z (2018) Discovering Latent Patterns of Urban Cultural Interactions in WeChat for Modern City Planning Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining Zhu D, Song D, Zhu B, Zhao J, Li Y, Zhang C, Zhu D, Yu C, Han T (2024) Understanding complex interactions between neighborhood environment and personal perception in affecting walking behavior of older adults: A random forest approach combined with human-machine adversarial framework. Cities 146. https://doi.org/10.1016/j.cities.2023.104737 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 28 Feb, 2026 Reviews received at journal 01 Jan, 2026 Reviews received at journal 17 Dec, 2025 Reviewers agreed at journal 08 Dec, 2025 Reviewers agreed at journal 08 Dec, 2025 Reviewers invited by journal 26 Nov, 2025 Editor invited by journal 12 Nov, 2025 Editor assigned by journal 11 Nov, 2025 Submission checks completed at journal 04 Nov, 2025 First submitted to journal 03 Nov, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8017731","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":552271609,"identity":"09dd7b48-2172-4ee9-b6d2-966c091a4f2f","order_by":0,"name":"Yiwei Mo","email":"","orcid":"","institution":"Hubei Institute of Fine Arts","correspondingAuthor":false,"prefix":"","firstName":"Yiwei","middleName":"","lastName":"Mo","suffix":""},{"id":552271611,"identity":"7020f430-97f5-458e-9a0a-6273eea78b89","order_by":1,"name":"Chao Luo","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyUlEQVRIiWNgGAWjYBACPiBmBmI5A2bGB1A2AcAGVWZswMxsQJqWxA0MRGuRSD74uaDGJn07OzPjB4YK68QG9rMHCGhJS5aecSwtd2czM7MEw5n0xAaevAQCWnLMmHnYDuduOMx/QIKx7XBigwSPARFa/h1ONzjMzPyD8R+xWnjbDicAtbBJMDYQo4XnWbI0b1+a4QagFouEY+nGbTw5+LXwswNDjOebjbzB+cPMNz7UWMv2s5/BrwUVJDBAYmoUjIJRMApGAYUAAIRyOUVkVoAFAAAAAElFTkSuQmCC","orcid":"","institution":"Hubei Institute of Fine Arts","correspondingAuthor":true,"prefix":"","firstName":"Chao","middleName":"","lastName":"Luo","suffix":""},{"id":552271613,"identity":"ec5b68d5-efd3-4d56-b60a-5ff07a592e1d","order_by":2,"name":"Youpeng Lu","email":"","orcid":"","institution":"Huazhong University of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Youpeng","middleName":"","lastName":"Lu","suffix":""},{"id":552271614,"identity":"eb1af6f8-03b2-45cd-8038-475ecd3d4d03","order_by":3,"name":"Qi Wu","email":"","orcid":"","institution":"Hubei Institute of Fine Arts","correspondingAuthor":false,"prefix":"","firstName":"Qi","middleName":"","lastName":"Wu","suffix":""},{"id":552271615,"identity":"b6d9ac70-7eb7-415b-9b6d-332b7088b498","order_by":4,"name":"Jingjing Wang","email":"","orcid":"","institution":"Wuhan University","correspondingAuthor":false,"prefix":"","firstName":"Jingjing","middleName":"","lastName":"Wang","suffix":""},{"id":552271616,"identity":"00e4af7d-6819-41f7-a553-4d6377a2102e","order_by":5,"name":"Zixiang Feng","email":"","orcid":"","institution":"Royal College of Art","correspondingAuthor":false,"prefix":"","firstName":"Zixiang","middleName":"","lastName":"Feng","suffix":""}],"badges":[],"createdAt":"2025-11-03 10:08:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8017731/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8017731/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":97132739,"identity":"a504d68e-711c-48ab-8b10-711efecffb8e","added_by":"auto","created_at":"2025-12-01 08:56:14","extension":"doc","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1589924,"visible":true,"origin":"","legend":"","description":"","filename":"final.doc","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/a435e2b6eb1549fc44062f0a.doc"},{"id":97132738,"identity":"3e3057f7-78b3-42ae-ace2-a31bca8f518d","added_by":"auto","created_at":"2025-12-01 08:56:14","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7492,"visible":true,"origin":"","legend":"","description":"","filename":"d9d2e051186e4e9895bc1327ab2cda41.json","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/4e3e50eac0af470f78d3bdb6.json"},{"id":97132746,"identity":"e3eb3647-bd95-4ea0-be51-b7c566fc914a","added_by":"auto","created_at":"2025-12-01 08:56:14","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":181252,"visible":true,"origin":"","legend":"","description":"","filename":"d9d2e051186e4e9895bc1327ab2cda411enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/0899923d971a3a4e11b084eb.xml"},{"id":97132748,"identity":"3ab85ecc-e9db-49a0-948d-fc1a7d100171","added_by":"auto","created_at":"2025-12-01 08:56:14","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":126961,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/46190b78b082fa6e44870973.png"},{"id":97132740,"identity":"42add800-29d7-4952-86df-03c10395875e","added_by":"auto","created_at":"2025-12-01 08:56:14","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":116344,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/e976b022d405e774c9a92592.png"},{"id":97142291,"identity":"86aec23d-8224-4823-a61d-6cb8e1d72e74","added_by":"auto","created_at":"2025-12-01 10:07:29","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":75115,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/680ae9ac9d19e5bf42f7d16b.png"},{"id":97142721,"identity":"abcc8fc3-6bfc-4f23-a1e4-0e1b21e8e80f","added_by":"auto","created_at":"2025-12-01 10:07:55","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":11683,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/ab6cfbd9ec1e90b3765c6bbc.png"},{"id":97132750,"identity":"d7a828df-f57b-49b8-ac22-4f166b821b17","added_by":"auto","created_at":"2025-12-01 08:56:14","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":79460,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/dc0583c802c5c37aff648801.png"},{"id":97142475,"identity":"652d8736-772e-4c47-8c8f-69f33e292940","added_by":"auto","created_at":"2025-12-01 10:07:38","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":82652,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/e33e7b37d9850c62a83820d2.png"},{"id":97132742,"identity":"183dcf95-502b-4b5c-82ef-a6601da77b6b","added_by":"auto","created_at":"2025-12-01 08:56:14","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":56265,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/93c8101d4c8a0a1a0c755084.png"},{"id":97132753,"identity":"9f1ac861-2550-494d-aca1-582924970d56","added_by":"auto","created_at":"2025-12-01 08:56:15","extension":"xml","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":179225,"visible":true,"origin":"","legend":"","description":"","filename":"d9d2e051186e4e9895bc1327ab2cda411structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/a1d342060322f82685fd3f8a.xml"},{"id":97132752,"identity":"ff174ee9-bfb8-4090-b244-eeb9583da10f","added_by":"auto","created_at":"2025-12-01 08:56:15","extension":"html","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":186418,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/40db9873364431a0b213ca7a.html"},{"id":97142190,"identity":"60c53616-b6ae-4811-84c8-72dac6358ba6","added_by":"auto","created_at":"2025-12-01 10:07:23","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":174866,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eResearch Framework.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/90eaf11ecfaf8368eaffffe4.jpeg"},{"id":97132737,"identity":"25a5cb80-b6d2-4d29-8064-9d014631c048","added_by":"auto","created_at":"2025-12-01 08:56:14","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":541182,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSpatial Distribution of POI Data.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/f22faf08f22d3b3de75e730e.jpeg"},{"id":97142357,"identity":"d1c5dbd0-6e7d-4636-8199-1f771fdc2c1b","added_by":"auto","created_at":"2025-12-01 10:07:33","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":449434,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eStudy Area.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/44e13fedc0b379922fb6f55f.png"},{"id":97132734,"identity":"7c8e2ff0-81a3-4d76-8f32-d2b075fe36f2","added_by":"auto","created_at":"2025-12-01 08:56:14","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":63044,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eRidge diagrams for different business types.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/cce919f9991054cf123270e4.png"},{"id":97141594,"identity":"e0e8cd8d-ab14-4b36-8e54-2301ae990afc","added_by":"auto","created_at":"2025-12-01 10:06:50","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":102354,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSummary chart of feature importance SHAP values. (a) CAT; (b) ISS; (c) LEX; (d) RTL.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/1d3dc184f5676d3b191a8238.jpeg"},{"id":97143008,"identity":"843ff0ae-0b73-4422-8272-7a65adb429fb","added_by":"auto","created_at":"2025-12-01 10:08:11","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":114229,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/0a39534a8ffe2cb7bdaa54b0.jpeg"},{"id":97142517,"identity":"193681aa-b019-4b02-9494-cac9838276d6","added_by":"auto","created_at":"2025-12-01 10:07:41","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":118784,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eInteraction Contribution Diagram of Driving Factors. \u003c/strong\u003eNote: a CAT: Interaction strengths.b ISS: Interaction strengths. c LEX: Interaction strengths d. RTL: Interaction strengths\u003c/p\u003e","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/60478b9dec0131cd9f0080b4.jpeg"},{"id":97145405,"identity":"bcfe303e-67bd-494d-8556-eb7c664f5a5c","added_by":"auto","created_at":"2025-12-01 10:13:52","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2669232,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8017731/v1/65853628-02f9-4245-9aef-46f130e36775.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"The Influence of Street-Scene Emotional Perception on Commercial Satisfaction: A Study Based on Random Forest Model and SHAP Algorithm","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eUrban streets are not only transportation networks connecting different functional zones but also constitute primary public spaces for social life through the spatial integration of roads, buildings, and associated service facilities (Mehta, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2013\u003c/span\u003e). Under the combined influence of streetscape characteristics and commercial functions, necessary, optional, and social activities such as walking, shopping, leisure, and social interaction naturally intertwine and transform, collectively shaping the attractiveness and vibrancy of urban streets (Istrate, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Whyte, \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e1980\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eGrounded in servicescape theory Bitner, an increasing number of interdisciplinary studies have demonstrated the associations between streetscape features, emotional perception, and commercial satisfaction (Han et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Koo et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). However, several limitations remain. First, many studies tend to overemphasize safety perception or focus on a single business category (e.g., catering services), often simplifying \"street commercial satisfaction\" into a homogeneous concept while overlooking its differentiated functional and emotional mechanisms. Empirical evidence, however, suggests that different business types\u0026mdash;such as catering, retail, cultural, and entertainment sectors (\u0026mdash;exhibit varying sensitivities to emotional cues (Correia \u0026amp; Kozak, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Meng et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), most previous research has primarily relied on linear modeling approaches, which may fail to capture potential nonlinear or threshold effects, as well as the interactive mechanisms between emotional perception and commercial functionality. Such methodological limitations may restrict the explanatory power and predictive robustness of existing models (Li et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Mittal \u0026amp; Kamakura, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2009\u003c/span\u003e) .\u003c/p\u003e\u003cp\u003eTo address these limitations, this study takes Wuhan, China, as a case to construct a multi-category analytical framework linking streetscape emotion and commercial satisfaction. Specifically, the study:(1) applied a deep learning model trained on the MIT Place Pulse dataset to extract six perceptual dimensions\u0026mdash;Aesthetics, Safety, Vibrancy, Wealth, Boring, and Depression\u0026mdash;from street-view images;(2) used point-of-interest (POI) data from Dianping to classify street-level commercial activities into four major business types\u0026mdash;Catering Services, Retail and Trade Locations, In-store Service Sectors, and Leisure and Experience Venues\u0026mdash;and employed consumer ratings as a proxy indicator of satisfaction; and(3) adopted the Random Forest (RF) model in combination with SHapley Additive exPlanations (SHAP) to quantify the relative importance and influence effects of emotional perception variables across different business types.\u003c/p\u003e\u003cp\u003eAt the theoretical level, the study integrates insights from servicescape theory (Bitner, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1992\u003c/span\u003e), public space vitality theory, and emotional geography (Mehta, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2013\u003c/span\u003e) to conceptualize the pathway from physical-space emotional perception to psychological cognition. At the practical level, by revealing the differentiated emotional drivers of satisfaction across business categories, this research provides empirical evidence for the design of emotionally responsive pedestrian environments in mixed-use streets, as well as actionable decision-support tools for street design and urban spatial governance.\u003c/p\u003e"},{"header":"2 Literature review","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1Reframing the Streetscape as an Open-Air Service Scape\u003c/h2\u003e\u003cp\u003eBitner's (1992) servicescape theory established the foundational S\u0026ndash;O\u0026ndash;R (Stimulus\u0026ndash;Organism\u0026ndash;Response) framework in environmental behavior research. It posits that the physical environment shapes individuals' cognitive and affective responses through atmospheric, spatial, and symbolic cues, thereby influencing behavioral outcomes such as satisfaction, approach, and avoidance. As a core theory in service marketing, traditional research has primarily focused on controllable indoor commercial environments\u0026mdash;such as stores and restaurants\u0026mdash;where managers are assumed to systematically manipulate environmental elements to optimize customer experience.\u003c/p\u003e\u003cp\u003eIn recent years, an increasing number of interdisciplinary studies have extended servicescape theory to the analysis of street environments and street-level commerce. This body of research demonstrates the applicability of the atmospheric concept to outdoor streetscapes: the adjacent streetscape constitutes an open-air service interface for street commerce (Whyte, \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e1980\u003c/span\u003e) Its architectural fa\u0026ccedil;ades, greenery, and pedestrian activities collectively form an open atmospheric system that reinforces the mutual visibility of \"seeing and being seen.\" These spatial and social cues act on consumers' emotional perception before, during, and after the service encounter, thereby affecting overall satisfaction (Chen et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Verhoef et al., \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2009\u003c/span\u003e) this expanded paradigm, the street environment is no longer a neutral backdrop but an active service space that stimulates users' agency through multisensory stimuli, informal interactions, and public visibility (Mehta, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2013\u003c/span\u003e). The permeability between indoor and outdoor boundaries challenges the conventional management-control assumptions of servicescape theory and calls for an extended understanding of \"distributed co-production\" within the service atmosphere (Gehl, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). Consequently, the modern streetscape has evolved into a hybrid socio-spatial system in which environmental design, behavioral expression, and emotional perception co-construct the quality of urban experience.\u003c/p\u003e\u003cp\u003eThis theoretical shift challenges the core premises of the original servicescape framework\u0026mdash;namely, the assumptions of clear environmental boundaries, unitary subjects, and centralized control\u0026mdash;and highlights the co-constructed and socially embedded nature of street atmospheres. Reconstructing an expanded framework capable of explaining emotional experience in outdoor, dynamic, and co-governed environments has thus become a central research focus (Whyte, \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e1980\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Emotional Perception Dimensions and Commercial Satisfaction\u003c/h2\u003e\u003cp\u003eWithin the servicescape theoretical framework, the physical environment primarily influences satisfaction and behavioral feedback through emotional perception. At the street scale, Lynch's (1960) concept of imageability reveals how street elements evoke collective atmospheres and shared memories((Lynch, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e1960\u003c/span\u003e), while Jacobs's (1961) metaphor of the \"sidewalk ballet\" underscores the role of everyday interaction in fostering community safety and identity(Jacobs, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e1961\u003c/span\u003e). The development of emotional geography further situates emotional experience within the relational and cultural fabric of place, conceptualizing the street as a dialectical unity of material and social processes (Mehta, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2013\u003c/span\u003e).Drawing on the Mehrabian\u0026ndash;Russell (1974) model, prior research has established an emotional\u0026ndash;behavioral chain of \"environmental stimuli \u0026rarr; affective state (e.g., safety, pleasure) \u0026rarr; satisfaction and approach behavior,\" bridging micro- (individual), meso- (street), and macro-level (urban) studies on spatial vitality (Foroudi et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shukri et al., \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) (Yi \u0026amp; Gim, \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). This paradigm shift from physical determinism to emotional mediation marks the transition of street design research on spatial vitality from correlation toward explanation.\u003c/p\u003e\u003cp\u003eHowever, existing studies often exhibit a form of typological blindness\u0026mdash;treating \"commercial satisfaction\" as a homogeneous construct and conducting analyses aggregated at the street or district level (Walsh et al., \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). Recent empirical findings indicate that perceived safety and walkable streetscape attributes associated with affective responses such as pleasure and aesthetics play a crucial role in shaping satisfaction within catering contexts (Koo et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) (Han et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Nevertheless, few studies have examined the streetscape as a servicescape influencing differentiated emotional pathways across distinct business types, including catering, retail, cultural, and entertainment sectors (Ho et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Stroebele \u0026amp; De Castro, \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2004\u003c/span\u003e) example, catering services are highly dependent on warmth and social comfort (Heung \u0026amp; Gu, 2012); retail environments rely on perceived order and safety (Correia \u0026amp; Kozak, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2016\u003c/span\u003e༉; while cultural and leisure spaces emphasize aesthetic pleasure and place identity (Dom\u0026iacute;nguez-Quintero et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). This differentiation highlights the need for streetscape emotion\u0026ndash;commercial satisfaction models to recognize the heterogeneity of business functions and to explore contextual variations in emotional influence mechanisms. Such inquiry contributes both theoretical and empirical foundations for refined placemaking and governance in multifunctional street environments.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Multi-Source Heterogeneous Data Analysis Supporting Large-Scale Urban Perception and Cognition Research\u003c/h2\u003e\u003cp\u003eThe rise of multi-source heterogeneous urban data has advanced large-scale research on urban perception and cognitive mechanisms. By integrating street-view imagery with point-of-interest (POI) data, scholars have been able to link objective dimensions of the built environment\u0026mdash;such as green-view index, enclosure, and fa\u0026ccedil;ade continuity\u0026mdash;and functional attributes including business type, density, and diversity (Zhang et al., \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) (Zhou et al., \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), with subjective dimensions of emotional perception (e.g., safety, beauty, vibrancy) and commercial satisfaction indicators such as online ratings (Yang \u0026amp; Zhang, \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) (Salesses, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2012\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eDeep learning models, particularly convolutional neural networks (CNNs), have enabled the automated extraction of emotional cues from visual data. The MIT Place Pulse project pioneered this approach by collecting global street-view images and crowdsourced perceptions of safety, beauty, and vibrancy (Naik et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Subsequent studies have trained deep neural networks on these datasets to generate large-scale spatial emotion maps, revealing significant correlations between perceived safety and socio-economic indicators such as crime rates and housing prices, and enhancing the reliability of cross-cultural emotion prediction (van Veghel et al., \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Wei et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eParallel to advances in visual analytics, diverse urban big data sources\u0026mdash;including social media reviews (Ashkezari-Toussi et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Despotovic \u0026amp; Hauser, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) (Zhou et al., \u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), pedestrian and location trajectories (Huang et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), mobile payment data (Feizizadeh et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), real-time pedestrian sensing (Cuesta-Mosquera et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Qin et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), and thermal heat maps (Du et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u0026mdash;have been widely employed to measure collective emotional atmosphere, functional cognition, and behavioral trends.\u003c/p\u003e\u003cp\u003eTogether, these data construct the structural characteristics of emotion and cognition under a bidirectional spatial\u0026ndash;behavioral mapping framework (Dubey, 2016; Luo et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), providing a quantitative foundation for examining the complex relationships between emotional perception and commercial satisfaction through large-scale data analysis.\u003c/p\u003e\u003cp\u003eIn terms of methodology, the Random Forest (RF) algorithm provides a robust tool for capturing nonlinear relationships and variable interactions often overlooked by traditional regression frameworks (Ma, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Zhu et al., \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Unlike Geographically Weighted Regression (GWR), which assumes spatial linearity, RF accommodates complex hierarchical structures and cross-variable effects while minimizing overfitting. The SHapley Additive exPlanations (SHAP) approach further enhances interpretability by quantifying each variable's marginal contribution to model output (Chen et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Luo et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Collectively, these methodological advances facilitate the identification of dominant emotional predictors and their industry-specific variations, providing a powerful analytical pathway to decode emotion\u0026ndash;satisfaction associations in mixed-use urban streetscapes.\u003c/p\u003e\u003c/div\u003e"},{"header":"3 Materials and Methods","content":"\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Research Framework\u003c/h2\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThis study followed an integrated five-stage framework encompassing data construction, sample balancing, model training, model interpretation, and result analysis (see Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eIn the data construction stage, the study selected the area enclosed by the Third Ring Road of Wuhan, Hubei Province, China, as the research boundary. Based on OpenStreetMap (OSM) street centerlines and Dianping point-of-interest (POI) data, POI types, ratings, and review counts were extracted within a 50 m buffer around each street centerline (Zhang et al., \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). All POIs were categorized into four macro-level commercial sectors following the GB/T 4754\u0026ndash;2017 and NAICS classification standards (Standardization Administration of \u0026amp; National Bureau of Statistics of, 2017), representing the primary business typologies along the streets. Simultaneously, the study combined the Place Pulse 2.0 dataset with Baidu Street View imagery (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://lbsyun.baidu.com\u003c/span\u003e\u003cspan address=\"https://lbsyun.baidu.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Using a pre-trained ResNet50 convolutional neural network, four-directional street-view images were processed to extract emotional perception scores across multiple affective dimensions at each sampling point. Each POI was then used as the center of a circular buffer with a 300 m radius\u0026mdash;approximately corresponding to a 3\u0026ndash;5 min walking distance (Koo et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). This spatial scale effectively covered the primary street segments in front of and behind catering venues and reflected the overall visual experience of visitors during their arrival and departure. Subsequently, all street-view images within each buffer were collected, and their perceptual features were averaged to represent the integrated emotional environment surrounding the corresponding POI (Han et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) this process, a fine-grained, multi-source dataset integrating streetscape emotional perception and commercial satisfaction was constructed, providing a robust empirical foundation for subsequent model analyses.\u003c/p\u003e\u003cp\u003eDuring the sample balancing stage, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to augment underrepresented categories, while random down-sampling was conducted on overrepresented ones. This ensured an equal number of samples across all categories, providing a balanced dataset for subsequent model training.\u003c/p\u003e\u003cp\u003eIn the model training stage, a Random Forest Regression (RF) model was employed, using six streetscape emotional perception indicators as independent variables and the commercial satisfaction rating as the dependent variable. Seventy-five percent of the samples were used for training and the remaining 25% for validation. To ensure robustness, the model underwent 100 iterations of random sampling and fitting. This approach allowed the exploration of nonlinear relationships and complex interaction effects embedded in the streetscape emotional data.\u003c/p\u003e\u003cp\u003eDuring the model interpretation stage, the study adopted the Tree SHAP algorithm and utilized multiple visualization techniques, including ridge density plots, permutation importance boxplots, partial dependence plots (PDP), interaction heatmaps, and SHAP beeswarm plots. These visual representations facilitated a detailed analysis of how each perception indicator influenced commercial satisfaction. The figures illustrated the relative importance of features, nonlinear response patterns, inter-feature interactions, and marginal contributions to model output, thereby providing a transparent visualization framework for understanding the complex association between streetscape emotional perception and commercial satisfaction.\u003c/p\u003e\u003cp\u003eThe result analysis was conducted across four dimensions:(1) describing the distribution of commercial satisfaction across different business types;(2) assessing the relative importance of six perception indicators;(3) revealing nonlinear responses of perception indicators through PDP analyses; and(4) quantifying and visualizing feature interactions to elucidate the multifaceted effects of streetscape emotional perception on commercial satisfaction.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Study Data\u003c/h2\u003e\u003cdiv id=\"Sec9\" class=\"Section3\"\u003e\u003ch2\u003e3.2.1 Point of Interest (POI) Data\u003c/h2\u003e\u003cp\u003eIn research on streetscape emotions, Points of Interest (POIs) serve as spatial representations of the \"third place\" beyond home (the first place) and work (the second place), encompassing named locations such as parks, caf\u0026eacute;s, and bookstores that accommodate social, leisure, and recreational activities (Psyllidis et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). With the advancement of digital technologies, POIs\u0026mdash;expressed through geographic coordinates\u0026mdash;have become proxy variables for real-world places and are widely applied in navigation, social activity analysis, and spatial research (Dhakal \u0026amp; Khadka, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). The rise of crowdsourced platforms such as Yelp, Foursquare, and OpenStreetMap (OSM) has further enriched POI diversity and attribute dimensions (e.g., operating hours, ratings, and photos), offering unprecedented spatial-semantic granularity for urban research (Gao et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eIn China, Dianping has emerged as one of the leading local lifestyle service platforms, featuring an extensive user base and rich POI data. Since its establishment in 2003, Dianping has rapidly evolved into a key digital infrastructure for urban life, covering POI information across sectors such as dining, entertainment, tourism, and retail (Qin et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Xu et al., \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). In this study, POI data were collected from Dianping within a 50 m buffer of street centerlines across the area enclosed by Wuhan's Third Ring Road(see Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The dataset included 17 primary categories and 221 secondary subcategories. Among these, four business types were identified as having significant associations with streetscape perceptual environments; therefore, the analysis focused on experience-oriented business types. Catering Services (CAT) include restaurants, caf\u0026eacute;s, and other food-related venues. Customer satisfaction in this sector is closely linked to physical settings, ambiance, and service quality (Wang et al., \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). This category was selected because its customer experience strongly depends on environmental cues and emotional engageme. Retailre visual components such as interior design and lighting directly shape emotional perception. Retail and Trade Locations (RTL) comprise supermarkets, home furnishing, and appliance stores\u0026mdash;spaces frequently visited in daily life. Satisfaction within this category is jointly influenced by product attributes and visual atmosphere, consistent with retail environment theories emphasizing the direct i. Inct of aservice sectorserchandising on consumer behavior (Luo \u0026amp; Wang, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; McKenzie \u0026amp; Romm, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).In-store Service Sectors (ISS) include businesses such as beauty salons and healthcare services (Kleeman et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Satisfaction in this category is primarily driven by accessibility and by the intensive interaction between customers and the physical setting during service delivery. T. Leisureunding streetscape exerts a notable influence on consumer decisions and experiential quality.Leisure and Experience Venues (LEX) encompass immersive entertainment spaces such as karaoke bars and family activity centers (Yang et al., \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Satisfaction in this category depends heavily on ambiance and social experience, with environmental elements like decoration and lighting playing a crucial role in shaping emotional engagement.\u003c/p\u003e\u003cp\u003eUnder the research objective of examining the influence mechanisms of external streetscape cues on adjacent commercial satisfaction, two types of POIs were excluded. First, function-oriented POIs (e.g., educational training centers, guesthouses, medical facilities) were removed because customer satisfaction in these sectors is primarily driven by professional quality and operational efficiency, with only marginal influence from streetscape characteristics. Second, nominal POIs that merely serve as place markers without any service-related experience were also excluded, as they have no substantive value for emotional perception modeling. After data screening and remapping, a total of 25,065 valid POI entries were retained. Following the GB/T 4754\u0026ndash;2017 and NAICS classification standards, the dataset was categorized into four macro-level business types: Catering Services (CAT, 39.74%), Retail and Trade Locations (RTL, 4.05%), In-store Service Sectors (ISS, 41.58%), and Leisure and Experience Venues (LEX, 14.63%).\u003c/p\u003e\u003cp\u003eGiven the pronounced imbalance in sample sizes among the four categories, random undersampling was applied to reduce overrepresented classes (He \u0026amp; Garcia, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2009\u003c/span\u003e), while the Synthetic Minority Oversampling Technique (SMOTE) was employed to augment underrepresented ones (Chawla et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2002\u003c/span\u003e). The final dataset contained 3,000 samples per category, ensuring uniform sample size across business types. This classification framework focuses on environmentally sensitive, experience-oriented business sectors. Through rigorous selection and resampling, the dataset achieved structural balance, providing a consistent and reliable input for subsequent Random Forest and Tree-SHAP analyses.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section3\"\u003e\u003ch2\u003e3.2.2 Street-view images of Wuhan\u003c/h2\u003e\u003cp\u003eStreet-view images of the study area were obtained from the Baidu Maps Static API (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://lbsyun.baidu.com/\u003c/span\u003e\u003cspan address=\"https://lbsyun.baidu.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). The platform has continuously archived street-view imagery since 2013, with many locations regularly updated. As of 2019, street-view samples were collected along the road network at 200 m intervals in four orientations\u0026mdash;0, 90, 180, and 270\u0026deg;\u0026mdash;yielding a total of 21,095 sampling points(see Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Consistent with the Place Pulse 2.0 dataset, each street-view image was uniformly resized to 400 \u0026times; 300 pixels to ensure standardized input dimensions for model processing.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section3\"\u003e\u003ch2\u003e3.2.3 The Place Pulse 2.0 dataset\u003c/h2\u003e\u003cp\u003eThe Place Pulse 2.0 dataset, developed by the Massachusetts Institute of Technology (MIT), is a crowdsourced platform designed to collect human perceptual evaluations of urban environments (Naik et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Through pairwise image comparisons conducted by online volunteers, the platform poses questions such as \"Which place looks safer?\" and aggregates large-scale perceptual data from street-view imagery. Compared with the first version, Place Pulse 2.0 substantially expands both the scope and depth of the dataset, encompassing 110,988 street-view images from 56 cities across 26 countries, and evaluates them along six perceptual dimensions: Safety, Aesthetics, Vibrancy, Wealth, Depression, and Boring.\u003c/p\u003e\u003cp\u003eIn this study, the Place Pulse 2.0 dataset provided the foundational data for the emotional perception model. Each street-view image was labeled according to its perceived score on these dimensions, calculated through pairwise comparison statistics. Specifically, the perceptual score of each image was derived from its relative win\u0026ndash;loss frequency across multiple comparisons. These scores were further standardized using a Q-score normalization process, yielding a continuous scale ranging from 0 to 10((Wei et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe dataset has played a significant role in various fields, particularly in urban planning (van Veghel et al., \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), environmental psychology(Qiu et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), and street-view perception studies (Liu et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Its extensive geographic coverage and multidimensional perceptual attributes make it a valuable resource for analyzing how urban environments influence human perception. By examining these perceptual data, researchers can better understand how specific visual characteristics of streetscapes shape public evaluations of safety, vibrancy, and related affective dimensions, thereby providing theoretical support for urban design and planning.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e3.3 Data Processing and Method\u003c/h2\u003e\u003cdiv id=\"Sec13\" class=\"Section3\"\u003e\u003ch2\u003e3.3.1 Mapping human perception of urban landscape using ResNet50\u003c/h2\u003e\u003cp\u003eIn deep learning, the depth of a neural network\u0026mdash;defined by the number of layers\u0026mdash;is critical to its performance. Theoretically, deeper networks possess stronger representational capacity and can learn more complex features. However, in practical training, increasing network depth often leads to two major challenges:(1) Vanishing/Exploding Gradients: During backpropagation, gradients may exponentially decay or grow as the number of layers increases, making deep networks difficult to train.(2) Degradation Problem: Even when gradient issues are mitigated (e.g., through Batch Normalization), empirical results show that simply deepening the network may increase both training and testing errors, resulting in performance degradation.\u003c/p\u003e\u003cp\u003eTo address these issues, He et al. (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) introduced skip connections, allowing the network to learn residual mappings rather than direct transformations.(He et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) Compared with conventional convolutional (Conv) or fully connected (FC) layers, the Residual Block design facilitates gradient propagation and improves optimization stability.\u003c/p\u003e\u003cp\u003eFor neural network\u0026ndash;based regression, both model architecture and data quality are essential. As often stated, data define the upper limit of a model, while algorithms merely approximate it. Neural networks essentially learn statistical patterns from large datasets. In supervised learning, training data provide: (1) input\u0026ndash;output correspondences that enable function approximation; (2) distributional information that helps the model understand diversity and latent structures; and (3) opportunities for feature extraction, allowing the model to automatically learn useful representations from variable combinations. In general, larger datasets tend to enhance model generalization, whereas low-quality data can significantly degrade performance. In supervised image regression tasks, labels must accurately reflect perceptual reality. Because visual qualities of photographs are difficult to describe textually, we used numerical scoring to enable neural representation, transforming image evaluation into a mathematical regression problem. However, label definition itself introduces subjectivity. To reduce bias, we manually annotated the data through pairwise image comparisons, assigning relative perceptual judgments that were later converted into numerical scores via algorithmic calibration. We thus reconstructed human perception representation as an image-based learning problem. First, street-view images with verified perceptual labels were converted into paired datasets with relative scores to build a standardized training set. Second, deep learning models were trained to extract features from labeled images. Finally, the trained model was applied to predict perceptual scores for new images within the study area. The resulting scores quantitatively represented human perception. Because correlations among perceptual dimensions were weak, each dimension was trained independently using six separate models.\u003c/p\u003e\u003cp\u003eWe conducted comparative experiments using AlexNet, VGG, and ResNet architectures. Results indicated that ResNet, with residual connections, significantly outperformed conventional CNNs of comparable depth in regression accuracy (Krizhevsky et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Simonyan \u0026amp; Zisserman, \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Accordingly, ResNet was selected for this regression task. Among ResNet18, ResNet50, and ResNet101, experiments showed that ResNet18 lacked sufficient representational capacity due to limited parameters, while ResNet101 required excessive computational resources and was prone to instability and overfitting with our dataset. Thus, ResNet50 was adopted as the optimal architecture. Seventy percent of the dataset was used for training, with 60% as the training subset and 10% as the validation subset. The remaining 30% was reserved for model evaluation. The network was initialized with pretrained ResNet50 weights and fine-tuned for regression. The pretrained model was fine-tuned with an initial learning rate (lr) set to 1e\u0026thinsp;\u0026minus;\u0026thinsp;6, which was dynamically adjusted during training. The original ResNet output layer, designed for classification, was retained, and a fully connected (FC) regression layer was appended to predict continuous values. The cross-entropy loss function was used as the training objective to stabilize gradient descent, while model validation and testing performance were assessed using the mean relative error between predicted and true labels. Training was performed on an NVIDIA RTX 2080Ti GPU with a batch size (batch_size) of 32 and a maximum of 100 epochs. Early stopping was applied when validation performance deteriorated or plateaued to prevent overfitting and improve generalization stability.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section3\"\u003e\u003ch2\u003e3.3.2 Random Forest regression\u003c/h2\u003e\u003cp\u003eRandom Forest (RF) regression is an ensemble learning method that fits multiple decision tree regression models on randomly sampled subsets of both data and features, and then aggregates their outputs by averaging (Breiman, 2001). Prior studies have demonstrated that RF regression can effectively capture nonlinear relationships among variables (Gao et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Therefore, this study also employed RF regression to examine the relationships between street-view emotional perception dimensions and commercial satisfaction, including the relative importance of variables and their partial dependence relationships. In this study, 75% of the samples were used for model training and the remaining 25% for validation. To ensure robustness, the random sampling and model fitting procedures were repeated 100 times, and the averaged outcomes were used for analysis. The coefficient of determination (R\u0026sup2;) and the root mean square error (RMSE) were adopted as performance evaluation metrics for the model.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section3\"\u003e\u003ch2\u003e3.3.3 SHapley Additive exPlanations\u003c/h2\u003e\u003cp\u003eAlthough Random Forest (RF) regression can effectively reveal nonlinear relationships among variables, it has certain limitations in capturing interaction effects, which are crucial in understanding the determinants of commercial satisfaction. The SHAP algorithm originates from the Shapley value concept in cooperative game theory, which aims to fairly distribute payoffs among participants in a coalition. In machine learning, SHAP assigns each feature a contribution score (i.e., SHAP value) that quantifies its influence on model outputs (Lundberg \u0026amp; Lee, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) SHAP, a variant of the SHAP algorithm, is particularly well-suited for tree-based models such as Random Forests. It further decomposes feature contributions into main effects and interaction effects (Lundberg et al., 2018; Ma, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Therefore, this study employed the Tree SHAP algorithm to detect the interactions among street-view emotional perception dimensions and their effects on commercial satisfaction, as well as to quantify the proportion of contribution attributable to feature interactions.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"4 Results","content":"\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003e4.1 Statistical Description of Commercial Satisfaction and Streetscape Perception\u003c/h2\u003e\u003cp\u003eDescriptive statistics of commercial satisfaction and streetscape perception were first computed for the four business types\u0026mdash;Catering Services (CAT), Retail and Trade Locations (RTL), In-store Service Sectors (ISS), and Leisure and Experience Venues (LEX) (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Results indicated that Aesthetics (Aes) showed relatively small mean variation across business types. Both CAT and RTL exhibited mean Aesthetics (Aes) scores above 4.85 with low standard deviations, suggesting consistent aesthetic evaluations among customers. For Safety (Saf), the mean score of CAT (3.06) was slightly higher than that of the other business types, indicating a more balanced perception of safety within catering environments. The mean Vibrancy (Vib) scores were similar across business types (around 3.4) but showed large standard deviations, suggesting substantial variation in perceived activity levels among respondents. Wealth (Wth) showed relatively high mean values (\u0026asymp;\u0026thinsp;5.15) and small standard deviations, implying stable and consistent perceptions across business types. Depression (Dpr) also displayed high mean values (\u0026asymp;\u0026thinsp;5.83) and low variability, reflecting a cross-type consistency in this perception. In contrast, Boring (Bro) had a lower mean of approximately 5.2 with larger standard deviations, indicating higher individual variability in the perception of boredom.\u003c/p\u003e\u003cp\u003eRegarding commercial satisfaction, significant differences were observed among business types. The CAT sector showed the highest mean satisfaction (M\u0026thinsp;=\u0026thinsp;3.74), exceeding that of the RTL sector (M\u0026thinsp;=\u0026thinsp;3.49). RTL also exhibited a larger standard deviation, suggesting greater heterogeneity in satisfaction across customers within this category.\u003c/p\u003e\u003cp\u003eAs illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the satisfaction distribution of CAT was more concentrated, with a sharp peak, indicating generally high and consistent satisfaction among customers. The ISS category displayed a flatter curve with a lower peak, suggesting a more dispersed evaluation pattern without a clear concentration. The LEX category showed a bimodal distribution, implying the presence of two distinct customer groups with divergent satisfaction levels. By contrast, the RTL category exhibited a wide and shallow distribution, suggesting both lower overall satisfaction and higher intra-group variability.\u003c/p\u003e\u003cp\u003eOverall, the results demonstrate observable fluctuations in both streetscape perception and commercial satisfaction across business types, particularly in dimensions such as aesthetics, safety, and boredom. These findings highlight the subjective and diverse nature of customer perception and suggest that street design and business configuration should adopt differentiated optimization strategies tailored to the perceptual characteristics of each commercial type.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eStatistical Description Across Different Business Models.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMacro / Metric\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCAT\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eISS\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLEX\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eRTL\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e\u003cp\u003e4.858\u0026thinsp;\u0026plusmn;\u0026thinsp;0.066\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e\u003cp\u003e4.850\u0026thinsp;\u0026plusmn;\u0026thinsp;0.067\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e\u003cp\u003e4.842\u0026thinsp;\u0026plusmn;\u0026thinsp;0.068\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e\u003cp\u003e4.858\u0026thinsp;\u0026plusmn;\u0026thinsp;0.068\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSaf\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e\u003cp\u003e3.065\u0026thinsp;\u0026plusmn;\u0026thinsp;0.089\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e\u003cp\u003e3.055\u0026thinsp;\u0026plusmn;\u0026thinsp;0.092\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e\u003cp\u003e3.053\u0026thinsp;\u0026plusmn;\u0026thinsp;0.090\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e\u003cp\u003e3.064\u0026thinsp;\u0026plusmn;\u0026thinsp;0.089\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVib\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e\u003cp\u003e3.400\u0026thinsp;\u0026plusmn;\u0026thinsp;0.120\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e\u003cp\u003e3.410\u0026thinsp;\u0026plusmn;\u0026thinsp;0.119\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e\u003cp\u003e3.409\u0026thinsp;\u0026plusmn;\u0026thinsp;0.120\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e\u003cp\u003e3.406\u0026thinsp;\u0026plusmn;\u0026thinsp;0.121\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWth\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e\u003cp\u003e5.155\u0026thinsp;\u0026plusmn;\u0026thinsp;0.086\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e\u003cp\u003e5.150\u0026thinsp;\u0026plusmn;\u0026thinsp;0.089\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e\u003cp\u003e5.149\u0026thinsp;\u0026plusmn;\u0026thinsp;0.089\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e\u003cp\u003e5.143\u0026thinsp;\u0026plusmn;\u0026thinsp;0.084\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDpr\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e\u003cp\u003e5.837\u0026thinsp;\u0026plusmn;\u0026thinsp;0.105\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e\u003cp\u003e5.836\u0026thinsp;\u0026plusmn;\u0026thinsp;0.109\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e\u003cp\u003e5.830\u0026thinsp;\u0026plusmn;\u0026thinsp;0.106\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e\u003cp\u003e5.833\u0026thinsp;\u0026plusmn;\u0026thinsp;0.105\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBro\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e\u003cp\u003e5.227\u0026thinsp;\u0026plusmn;\u0026thinsp;0.116\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e\u003cp\u003e5.218\u0026thinsp;\u0026plusmn;\u0026thinsp;0.122\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e\u003cp\u003e5.217\u0026thinsp;\u0026plusmn;\u0026thinsp;0.127\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e\u003cp\u003e5.213\u0026thinsp;\u0026plusmn;\u0026thinsp;0.117\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eStar rating\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e\u003cp\u003e3.739\u0026thinsp;\u0026plusmn;\u0026thinsp;0.347\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e\u003cp\u003e3.717\u0026thinsp;\u0026plusmn;\u0026thinsp;0.457\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e\u003cp\u003e4.016\u0026thinsp;\u0026plusmn;\u0026thinsp;0.462\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e\u003cp\u003e3.490\u0026thinsp;\u0026plusmn;\u0026thinsp;0.444\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003e4.2 Relative importance analysis\u003c/h2\u003e\u003cp\u003eTo enhance the robustness and reproducibility of the findings, beyond the relative importance ranking illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, this study further examined the contribution patterns of each emotional perception dimension across different business types using SHAP distribution analysis.\u003c/p\u003e\u003cp\u003eOverall, the density peaks (density_peak) of all four business categories slightly fell on the left side of the zero axis (approximately \u0026minus;\u0026thinsp;0.05 to \u0026minus;\u0026thinsp;0.01), with all skewness values being positive. This distributional pattern indicates a shared tendency characterized by a near-zero or mildly negative baseline contribution coupled with a distinctly elongated positive tail. The combination of a \"slightly negative peak with positive skewness\" suggests that emotional dimensions generally exert mild net effects on satisfaction for most samples but exhibit asymmetric positive surges in certain contexts, thereby amplifying their overall explanatory power.\u003c/p\u003e\u003cp\u003eIn Catering Services (CAT), Depression (Dpr) and Vibrancy (Vib) demonstrated the strongest individual heterogeneity, with coefficients of variation (CV) of 156.13 and 183.46, respectively, both markedly higher than those of other dimensions. Their interquartile ranges (IQR) were also among the largest (Dpr\u0026thinsp;=\u0026thinsp;0.046; Vib\u0026thinsp;=\u0026thinsp;0.045). In contrast, Aesthetics (Aes) and Safety (Saf) exhibited much lower CVs (30.56 and 58.73) and narrower IQRs (0.039 and 0.043), implying greater stability in their contribution directions. All five dimensions showed positive skewness (e.g., Aes and Saf\u0026thinsp;\u0026asymp;\u0026thinsp;1.83), which, together with the mildly negative density peaks, formed a \"moderate central tendency with right-tail enhancement\" pattern.\u003c/p\u003e\u003cp\u003eIn In-store Service Sectors (ISS), Aesthetics displayed the most pronounced distributional fluctuation, with an IQR of 0.104, far exceeding other dimensions (Wealth\u0026thinsp;=\u0026thinsp;0.080; Safety\u0026thinsp;=\u0026thinsp;0.055; Vibrancy\u0026thinsp;=\u0026thinsp;0.052; Depression\u0026thinsp;=\u0026thinsp;0.052), and a skewness of 1.37, suggesting significant bidirectional expansion and right-tail amplification. Meanwhile, Aesthetics and Wealth had relatively low CVs (19.55 and 24.78), indicating a \"highly volatile but internally consistent\" pattern. In contrast, Vibrancy showed a high CV (94.36), suggesting that although its overall contribution remained close to the zero axis, considerable fluctuation occurred at specific points. The safety distribution was more stable (IQR\u0026thinsp;=\u0026thinsp;0.055; skewness\u0026thinsp;=\u0026thinsp;0.42), with values concentrated in the non-negative range.\u003c/p\u003e\u003cp\u003eIn Leisure and Experience Venues (LEX), Aesthetics again showed the greatest volatility (IQR\u0026thinsp;=\u0026thinsp;0.089; skewness\u0026thinsp;=\u0026thinsp;1.59), followed by Safety and Depression (IQR\u0026thinsp;=\u0026thinsp;0.078 and 0.075, respectively). Vibrancy had the smallest IQR (0.068) but a relatively high CV (69.51), indicating that while its overall effect hovered near zero, a certain degree of heterogeneity persisted. The density peaks of all dimensions remained mildly negative (approximately \u0026minus;\u0026thinsp;0.026 to \u0026minus;\u0026thinsp;0.019), which, combined with positive skewness, yielded a \"broadly bidirectional but right-tail dominant\" distributional pattern.\u003c/p\u003e\u003cp\u003eIn Retail and Trade Locations (RTL), Aesthetics exhibited the highest volatility and heterogeneity across all samples, with an IQR of 0.091, a CV of 266.91, and a skewness of 2.44, revealing pronounced right-tail amplification and inter-sample variability. Vibrancy also showed wide distributional spread (IQR\u0026thinsp;=\u0026thinsp;0.077; skewness\u0026thinsp;=\u0026thinsp;2.31). Although Safety and Wealth exhibited relatively narrow IQRs (0.054 and 0.052), their extreme right-tail skewness values (5.17 and 5.49, respectively) suggested the presence of rare but exceptionally strong positive enhancements. Depression occupied a moderate position (IQR\u0026thinsp;=\u0026thinsp;0.069; CV\u0026thinsp;=\u0026thinsp;104.73; skewness\u0026thinsp;=\u0026thinsp;3.04).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\u003ch2\u003e4.3 Dependency between Driving Factors and Commercial Satisfaction\u003c/h2\u003e\u003cp\u003ePartial dependence plots (PDPs) revealed nonlinear relationships between each emotional perception dimension and commercial satisfaction (see Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). Results are reported by business type.\u003c/p\u003e\u003cp\u003eIn the catering services sector, Aesthetics (Aes) exhibited a nonlinear positive relationship with commercial satisfaction. As Aes approached 4.8, satisfaction began to increase gradually, and a sharp rise occurred between approximately 4.9 and 5.0, indicating a clear threshold effect in the influence of aesthetics on satisfaction. Vibrancy (Vib) showed a positive relationship with satisfaction; satisfaction increased steadily from 3.2 to 3.4, and the rate of increase accelerated near 3.4, indicating a threshold-like acceleration. Wealth (Wth) displayed nonlinear variation: satisfaction decreased between 5.0 and 5.2, yet rose sharply as Wth approached 5.2, evidencing a threshold effect. Depression (Dpr) also varied nonlinearly, with pronounced fluctuations in the 5.6\u0026ndash;5.8 interval. A value of 5.7 was identified as a critical threshold: beyond this point satisfaction declined rapidly and became more volatile, after which the downward trend moderated but remained at a lower level.\u003c/p\u003e\u003cp\u003eIn the in-store service sectors sector, Aesthetics (Aes) exhibited marked nonlinear volatility. As Aes approached 4.8\u0026ndash;4.9, satisfaction increased sharply and then entered a period of pronounced fluctuation; in particular, satisfaction dropped notably near 4.9 before rebounding, identifying 4.9 as a key threshold. Safety (Saf) showed a comparatively stable relationship with satisfaction: within 3.0\u0026ndash;3.2 satisfaction varied little and no clear threshold effect was observed. Vibrancy (Vib) displayed a positive, nonlinear relationship: satisfaction rose steadily from 3.2 to 3.4, with a notable acceleration at 3.4, indicating 3.4 as a potential threshold where increases in Vib produce accelerated gains in satisfaction. Wealth (Wth) showed nonlinear change: from 5.0 to 5.2 satisfaction rose gradually, yet the rate increased sharply near 5.2, suggesting 5.2 as a critical threshold. Depression (Dpr) exhibited pronounced nonlinear variation in the 5.8\u0026ndash;6.0 interval; satisfaction fell precipitously near 5.9\u0026ndash;6.0, identifying 5.9\u0026ndash;6.0 as critical thresholds. Boring (Bro) also varied nonlinearly: stability prevailed between 5.0 and 5.2, but significant volatility occurred approaching 5.2, and extreme variation appeared between 5.3 and 5.4. The 5.2\u0026ndash;5.3 range was thus identified as a key threshold at which sustained, substantial changes in satisfaction occur.\u003c/p\u003e\u003cp\u003eIn the leisure and experience venues sector, Aesthetics (Aes) produced strong nonlinear effects: satisfaction was relatively stable from 4.8 to 4.9, but increased sharply from 4.9 to 5.0, with a pronounced jump near 4.9, indicating high sensitivity to aesthetic changes. Safety (Saf) showed limited fluctuation, primarily within 3.0\u0026ndash;3.2, with small changes in satisfaction. Vibrancy (Vib) had only minor variation, concentrated in 3.2\u0026ndash;3.4, implying a limited effect on satisfaction. Wealth (Wth) was positively associated with satisfaction: increases from 5.0 to 5.1 produced modest gains, whereas the rate of increase accelerated from 5.1 to 5.2, with 5.2 identified as a key threshold. Depression (Dpr) produced large satisfaction fluctuations in 5.8\u0026ndash;6.0, with a sharp decline near 6.0; 5.9\u0026ndash;6.0 were therefore threshold values. After an abrupt rise followed by a slight fall, the net effect was a weak positive association overall. Boring (Bro) was stable between 5.0 and 5.2, but showed significant volatility near 5.2 and a pronounced jump at 5.4; 5.2\u0026ndash;5.3 was identified as the threshold range.\u003c/p\u003e\u003cp\u003eIn the retail and trade locations sector, Aesthetics (Aes) showed a marked negative association with satisfaction. While changes were modest near 4.8\u0026ndash;4.9, satisfaction varied dramatically between 4.9 and 5.0, with a pronounced decline near 4.9, indicating 4.9 as a critical threshold beyond which fluctuations intensified. Safety (Saf) related positively to satisfaction: as Saf increased from 3.0 to 3.2, satisfaction gains accelerated, with 3.2 serving as a key threshold beyond which improvements in Saf produced faster satisfaction increases. Vibrancy (Vib) exhibited nonlinear volatility, particularly within 3.2\u0026ndash;3.6, including multiple local peaks. Wealth (Wth) showed noticeable fluctuation in 5.0\u0026ndash;5.2; near 5.1\u0026ndash;5.2 satisfaction variation increased, and a rapid rise occurred approaching 5.2, although it did not return to the highest observed level. Depression (Dpr) displayed nonlinear fluctuation in 5.8\u0026ndash;6.0, with a sharp decline near 6.0; 5.9\u0026ndash;6.0 were identified as threshold values, and exceeding this range produced large increases in satisfaction variability. Boring (Bro) showed nonlinear changes, notably in 5.2\u0026ndash;5.4; 5.3 emerged as a key threshold where satisfaction volatility began to increase, and near 5.4 the upward change in satisfaction moderated and ultimately declined.\u003c/p\u003e\u003cp\u003eIn summary, all four business categories exhibited distinct perceptual threshold ranges, among which the threshold effects of Aesthetics, Safety, and Wealth were the most pronounced. This finding suggests that emotional cues in streetscapes shape differentiated pathways influencing satisfaction across various commercial types.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\u003ch2\u003e4.4 Interaction Effects of Driving Factors on Commercial Satisfaction\u003c/h2\u003e\u003cp\u003eBased on the SHAP interaction analysis, the interactional contributions among emotional dimensions exhibited notable variations across different business categories. In the Catering Services (CAT) sector (see Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea), Boring (Bro) demonstrated the strongest interaction intensity with other perceptual factors. The SHAP value approached 0.008, particularly in its interaction with Wealth (Wth) and Vibrancy (Vib), where commercial satisfaction fluctuated markedly. This indicates that social engagement exerts the most salient influence among all interactions. Strong interaction effects were also observed between Safety (Saf) and Depression (Dpr), Dpr and Vib, as well as wealth and Aesthetics (Aes), with SHAP values ranging from approximately 0.0075 to 0.008, suggesting that these interactions play an important role in explaining variations in satisfaction.\u003c/p\u003e\u003cp\u003eIn the In-store Service Sectors (ISS) category (see Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb), the interaction between Wealth (Wth) and Aesthetics (Aes) showed the greatest intensity, with a SHAP value of 0.014, indicating that their joint effect had the most substantial influence on satisfaction. Additionally, interactions between Aesthetics (Aes) and Depression (Dpr), as well as between Aesthetics (Aes) and Boring (Bro), also demonstrated strong effects, with SHAP values of approximately 0.012, further emphasizing the centrality of aesthetics among perceptual dimensions.\u003c/p\u003e\u003cp\u003eIn the Leisure and Experience Venues (LEX) sector (see Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ec), the interactions between Aesthetics (Aes) and Vibrancy (Vib), and between Aesthetics (Aes) and Depression (Dpr), exhibited relatively high impacts, both with SHAP values exceeding 0.011. This suggests that these pairings significantly affect commercial satisfaction. Meanwhile, the interactions between Depression (Dpr) and Safety (Saf), Safety (Saf) and Wealth (Wth), Aesthetics (Aes) and Safety (Saf), as well as Boring (Bro) and Safety (Saf), also displayed strong effects, with SHAP values around 0.010, indicating that these combinations meaningfully influence satisfaction levels.\u003c/p\u003e\u003cp\u003eIn the Retail and Trade Locations (RTL) category (see Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ed), the interactions between Aesthetics (Aes) and Vibrancy (Vib), and between Aesthetics (Aes) and Depression (Dpr), exhibited the strongest intensities, with SHAP values exceeding 0.016. These combinations therefore exerted the most significant effects on satisfaction. Moreover, the interactions between Depression (Dpr) and Vibrancy (Vib), Boring (Bro) and Aesthetics (Aes), and Boring (Bro) and Wealth (Wth) also showed relatively strong influences, with SHAP values around 0.014, suggesting that these pairings contributed meaningfully to the observed variation in commercial satisfaction.\u003c/p\u003e\u003cp\u003eOverall, Aesthetics, Safety, and Vibrancy consistently functioned as key driving dimensions across all business categories. Their influence pathways were not isolated or linear but were jointly shaped through the synergistic interplay of multidimensional perceptual cues that together defined the overall experience of commercial satisfaction.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"5 Discussion","content":"\u003cdiv id=\"Sec22\" class=\"Section2\"\u003e\u003ch2\u003e5.1Emotion-Driven Patterns across Business Categories\u003c/h2\u003e\u003cp\u003eThis study employed a Random Forest model combined with SHAP interpretability analysis to systematically reveal, for the first time, the differentiated response mechanisms of multidimensional streetscape emotional perceptions on commercial satisfaction across distinct business categories, thereby overcoming the current research limitations of single-type and single-dimension analyses (Walsh et al., \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2011\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe Catering Services (CAT) sector demonstrated a distinctive \"Depression-dominated and Vibrancy-synergized\" pattern. Depression (Dpr) emerged as the most critical predictor, exhibiting a clear threshold effect on satisfaction: when Dpr scores exceeded 5.7, satisfaction declined sharply. This aligns with previous evidence that environmental stress significantly impairs dining experiences (Stroebele \u0026amp; De Castro, \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2004\u003c/span\u003e), while the present study quantitatively identifies this perceptual turning point. Notably, Vibrancy (Vib), though the second most important predictor, showed a robust positive influence, particularly when Vib\u0026thinsp;\u0026gt;\u0026thinsp;3.4, producing a marked enhancement effect. This finding supports Whyte's (1980) \"outdoor interface\" theory(Whyte, \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e1980\u003c/span\u003e༉, suggesting that catering environments rely on street vitality to create \"watchability\" yet must balance stimulation with pressure control to maintain comfort.\u003c/p\u003e\u003cp\u003eThe In-store Service (ISS) and Leisure and Experience Venues (LEX) sectors shared a \"Aesthetics-centered driving\" mechanism. In both categories, Aesthetics (Aes) ranked highest in SHAP importance, and its impact displayed pronounced bidirectionality: high Aes scores (Aes\u0026thinsp;\u0026gt;\u0026thinsp;4.9) strongly elevated satisfaction, whereas low aesthetic values triggered sharp declines. This supports environmental psychology perspectives, which posit that in high-contact service settings such as salons and caf\u0026eacute;s, visual aesthetics directly shape consumer identity and self-congruity (Japutra et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). However, a key difference emerged: in the LEX sector, Depression (Dpr) ranked fourth in importance\u0026mdash;significantly higher than in the ISS sector, where it was the least influential. The partial dependence plot (PDP) indicated that when Dpr\u0026thinsp;\u0026gt;\u0026thinsp;5.9, satisfaction dropped rapidly, suggesting that immersive leisure environments are more sensitive to environmental stress, whereas functional service contexts show greater tolerance.\u003c/p\u003e\u003cp\u003eThe Retail and Trade Locations (RTL) sector exhibited an \"aesthetic paradox.\" Although Aesthetics (Aes) ranked as the most important variable, its SHAP value distribution showed a pronounced negative skew, and the PDP revealed a clear satisfaction decline when Aes\u0026thinsp;\u0026gt;\u0026thinsp;4.9. This finding runs counter to conventional assumptions but is consistent with visual complexity theory: moderate aesthetic richness enhances attention and pleasure, whereas excessive aesthetic or visual complexity may surpass a cognitive threshold, increasing cognitive load, reducing processing fluency, and elevating mismatched expectations, ultimately decreasing satisfaction (Reber et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2004\u003c/span\u003e). Empirical research on retail spaces likewise shows that high visual complexity often reduces pleasure among low-engagement consumers and, through emotional mediation, diminishes store entry and purchase intentions. Hence, the intuitive notion that \"more aesthetics equals better experience\" does not necessarily hold true and may exhibit threshold effects (Jang et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Moreover, this study observed a notable satisfaction surge when Vibrancy (Vib) exceeded 3.4, suggesting that retail experiences rely more heavily on environmental dynamism and sensory stimulation to evoke immediate emotions and impulsive purchasing behaviors (Huang et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Therefore, in RTL contexts, enhancing static aesthetics without integrating dynamic vitality may inadvertently suppress overall satisfaction.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec23\" class=\"Section2\"\u003e\u003ch2\u003e5.2Business-Type Differentiation Mechanisms of Interaction Effects\u003c/h2\u003e\u003cp\u003eInteraction analysis further revealed the cross-dimensional emotional synergy mechanisms within each business category. In the Catering Services (CAT) sector, the interaction between Boring (Bro) and Wealth (Wth) was the strongest (SHAP\u0026thinsp;=\u0026thinsp;0.008). High-wealth environments, such as premium dining districts, appeared to mitigate the negative impact of boredom through symbolic consumption. When Wth exceeded 5.2, visual markers such as luxury storefronts and branded fa\u0026ccedil;ades functioned as \"anchors of identity recognition,\" effectively reducing customers' sense of boredom. Conversely, in lower-wealth contexts such as mass food streets, a \"density reinforcement\" effect likely emerged, creating a lively and vibrant atmosphere (\"renao\") that offset monotony. This outcome may relate to social interaction patterns embedded in street activities; however, as cultural orientation was not directly measured in this study, cross-cultural data are required for further validation.\u003c/p\u003e\u003cp\u003eIn the In-store Services (ISS) sector, a strong interaction between Aesthetics (Aes) and Wealth (Wth) (SHAP\u0026thinsp;=\u0026thinsp;0.014) was identified, reflecting a mechanism of signaling and credibility construction. The combination of high aesthetic design and refined facilities reduced uncertainty about service quality through dual \"aesthetic\u0026ndash;capital\" cues, particularly in functional service venues such as hair salons and repair shops. However, under lower Wth conditions, excessive aesthetic design appeared to induce cognitive dissonance, suggesting that inconsistency between exterior design and perceived prosperity may trigger psychological conflict. Given that consumers typically spend short durations in such venues and rely heavily on visual information for rapid quality assessments, the synergy between aesthetics and wealth likely generates a multiplier effect in visual evaluation.\u003c/p\u003e\u003cp\u003eIn the Leisure and Experience Venues (LEX) sector, nonlinear interactions between Aesthetics (Aes) and Vibrancy (Vib) (SHAP\u0026thinsp;=\u0026thinsp;0.011) were found to determine the degree of immersive experience. In highly aesthetic environments, customer engagement with visual stimuli intensified; however, when Vib surpassed a certain threshold, external stimulation began to compete for attentional resources, thereby undermining immersion. This finding aligns with the attentional resource competition model (Mehrabian \u0026amp; Russell, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e1974\u003c/span\u003e), suggesting that overstimulation may counteract the intended experiential benefits of aesthetic investment.\u003c/p\u003e\u003cp\u003eIn the Retail and Trade Locations (RTL) sector, a significant negative interaction between Aesthetics (Aes) and Depression (Dpr) was observed (SHAP\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.016). In high-depression environments, high aesthetic levels amplified the sense of spatial oppression through a \"cognitive contrast enhancement\" effect. This was particularly evident in luxury malls, where refined decorations and crowded traffic jointly evoked consumer anxiety, supporting the notion of an \"aesthetic oppression paradox.\" Conversely, moderate visual clutter in traditional markets appeared to reduce sensitivity to oppressive cues, suggesting a cultural tolerance for mild disorder in such contexts. These findings highlight the importance of balancing attractiveness and comfort in retail design: excessive aesthetic refinement may provoke negative emotional responses and consequently diminish overall commercial satisfaction.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec24\" class=\"Section2\"\u003e\u003ch2\u003e5.3Theoretical Reconstruction and Contextual Innovation\u003c/h2\u003e\u003cp\u003eGrounded in servicescape theory and emotional geography, this study extends existing theoretical frameworks in three primary ways. First, it empirically validates an \"open-air extension\" of servicescape theory. Although the original model primarily emphasized environmental control within enclosed service settings, the present findings indicate that outdoor streetscapes possess distinctive emotional dynamics. While Safety (Saf) significantly affects satisfaction, socially oriented dimensions\u0026mdash;particularly Vibrancy (Vib) and Boring (Bro)\u0026mdash;exert even greater influence. This pattern supports the \"street as open-air theater\" concept (Mehta, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2013\u003c/span\u003e), in which informal social interactions (e.g., street vending, casual neighbor exchanges) constitute core mechanisms of affective generation. The results therefore suggest that managers of open urban environments should prioritize fostering harmonious social interactions rather than focusing solely on the control of physical spatial design.\u003c/p\u003e\u003cp\u003eSecond, the study empirically confirms the \"emotion \u0026rarr; satisfaction\" transmission pathway proposed in emotional geography and simultaneously uncovers sectoral heterogeneity (Y\u0026uuml;ksel et al., \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). Specifically, in the Catering Services (CAT) sector, emotional dimensions such as Depression (Dpr) and Vibrancy (Vib) significantly influenced customer satisfaction. In contrast, in the Retail and Trade Locations (RTL) sector, Aesthetics (Aes) emerged as the dominant determinant. These findings imply that different business types may exhibit distinct moderating mechanisms within their emotional influence pathways, warranting further investigation through multi-group or moderated structural equation modeling (SEM).\u003c/p\u003e\u003cp\u003eThird, the results challenge two prevalent theoretical simplifications:\u003c/p\u003e\u003cp\u003e(1) The \"vitality supremacy\" assumption. While Ewing et al. (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2006\u003c/span\u003e) emphasized Vibrancy (Vib) as a core determinant of environmental appeal, the current findings reveal that its importance is highly contingent on business type((Ewing et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). In the Catering Services (CAT) sector, Vib ranked second in predictive importance, whereas in the Leisure and Experience Venues (LEX) sector, it ranked last and exhibited an upper threshold beyond which satisfaction declined. This suggests that urban design should aim for a balanced emotional composition rather than a singular pursuit of vitality.\u003c/p\u003e\u003cp\u003e(2) The linearity assumption. Traditional OLS-based models have often posited a simple positive linear relationship between safety (Saf) and satisfaction (Jani \u0026amp; Han, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). However, the Partial Dependence Plots (PDPs) in this study reveal a nonlinear association between the two. For instance, in the Catering Services (CAT) sector, when Saf scores were below 3.2, their impact on satisfaction was negligible, but once Saf exceeded this threshold, the marginal effect increased sharply. A similar S-shaped pattern was also observed in the Retail and Trade Locations (RTL) sector, further demonstrating the generalizability of nonlinear relationships in affective perception research.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec25\" class=\"Section2\"\u003e\u003ch2\u003e5.4 Emotion-Customized Spatial Intervention Strategies\u003c/h2\u003e\u003cp\u003eBuilding upon the results of the Random Forest model and SHAP interpretation, this study proposes an \"Emotion-Customized\" spatial intervention framework to support precision-based design strategies across different commercial environments. The findings demonstrate that the dominant emotional drivers vary significantly by business category. Therefore, urban design and commercial renovation should avoid applying uniform templates and instead adopt differentiated interventions according to the primary affective mechanisms of each sector.\u003c/p\u003e\u003cp\u003eIn the Catering Services (CAT) environment, model results indicated that Depression (Dpr) exhibited a clear inflection range between approximately 5.5 and 5.8, beyond which satisfaction declined sharply. This empirical threshold suggests the need for \"stress-reducing\" spatial interventions. Design strategies may focus on mitigating spatial enclosure and visual crowding by optimizing sight corridors, enhancing transparency of outdoor seating interfaces, and introducing localized greenery and natural lighting to foster a sense of visible relaxation and openness.\u003c/p\u003e\u003cp\u003eIn the In-store Services (ISS) and Leisure and Experience Venues (LEX) sectors, Aesthetics (Aes) and Wealth (Wth) displayed a synergistic relationship. When both dimensions reached higher levels, satisfaction improved markedly; however, excessive aesthetic refinement within low-wealth environments produced cognitive dissonance. Accordingly, design approaches should emphasize credibility through calibrated precision of visual and material details while avoiding ornamental excess. In the LEX sector specifically, Vibrancy (Vib) exhibited a nonlinear boundary within the 3.2\u0026ndash;3.6 range, suggesting the necessity of controlling the upper limit of external stimulation. Techniques such as zoned acoustic and lighting modulation or buffered circulation pathways may balance immersive engagement with ambient vitality.\u003c/p\u003e\u003cp\u003eIn the Retail and Trade Locations (RTL) environment, the model identified a negative interaction between high Aesthetics (Aes) and high Depression (Dpr), where satisfaction declined substantially when both values were elevated. This pattern reveals a phenomenon of \"aesthetic oppression\". Consequently, design priorities should shift from aesthetic intensification toward behavioral comfort enhancement, restoring a sense of everyday liveliness through optimized fa\u0026ccedil;ade openness, rhythm-controlled pedestrian flow, and tolerance for moderate visual irregularity within marketplace layouts.\u003c/p\u003e\u003cp\u003eOverall, emotion-customized design underscores contextual differentiation rather than aesthetic uniformity. By translating data-driven emotional thresholds into actionable design parameters, this research bridges machine learning analytics with urban design practice, providing empirical grounding and theoretical advancement for data-informed affective urbanism.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec26\" class=\"Section2\"\u003e\u003ch2\u003e5.5 Limitations and prospects\u003c/h2\u003e\u003cp\u003eThis study has several methodological limitations.(1) Static imagery bias: The analysis relied on static Baidu street-view images collected in 2019, which excluded temporal variations. For instance, nighttime economic periods (18:00\u0026ndash;24:00) with changing lighting conditions and vendor density, as well as seasonal transformations such as cherry blossom or rainy seasons, may substantially alter streetscape semantics but were not incorporated. To address this limitation, future research may integrate satellite-based nighttime light data with nocturnal street-view imagery to construct a spatiotemporal cube model, enabling the capture of dynamic changes in urban streetscapes.(2) Limited sensory modality: The study focused solely on the visual dimension, overlooking auditory factors (e.g., street vendor calls, traffic noise) and olfactory cues (e.g., food aroma, pollution) that likely affect emotional perception. Future efforts may deploy portable sensor networks to record multisensory data and combine these with social media sentiment analysis to build a multimodal emotional mapping framework, thereby achieving a more comprehensive understanding of how streetscapes shape emotional responses.(3) Black-box limitation of nonlinear mechanisms: Although the Random Forest model effectively captured threshold effects, it did not decompose the psychological pathways through which emotional perception influences satisfaction\u0026mdash;such as place attachment or consumption-related affect. Subsequent studies could integrate Bayesian Networks (BN) with Structural Equation Modeling (SEM) to quantify the transmission weights along the emotion\u0026ndash;function cognition pathway and further elucidate the complexity of underlying nonlinear mechanisms.(4) Cross-cultural adaptability of perceptual models: Cultural perception bias may influence ratings of aesthetics and safety. Therefore, model fine-tuning tailored to local linguistic and semantic contexts is necessary to ensure contextual validity.(5) User-generated content bias: Review data from the Dianping platform may be subject to selective self-reporting bias. Future research could integrate check-in or transaction data to enhance reliability and validity. Collectively, these methodological limitations provide valuable directions for future refinement and may contribute to a more precise understanding of the multidimensional mechanisms underlying emotional perception of urban streetscapes.\u003c/p\u003e\u003c/div\u003e"},{"header":"6 Conclusions","content":"\u003cp\u003eThis study extends the theory of service scape to open street environments by integrating street-view imagery, POI data, and machine learning methods. It systematically reveals the multidimensional mechanisms by which emotional perception of streetscapes influences commercial satisfaction and demonstrates differentiated patterns across business types.\u003c/p\u003e\u003cp\u003eFirst, the findings identify business-type-specific emotional driving patterns. The Catering Services (CAT) sector exhibits a \"Depression-dominated and Vibrancy-synergistic\" mode, in which Depression (Dpr) serves as the primary negative predictor with a distinct threshold effect\u0026mdash;once exceeding a critical value, satisfaction declines sharply. In contrast, Vibrancy (Vib) exerts a positive synergistic influence, particularly when surpassing a moderate level, leading to a marked increase in satisfaction. Both the In-store Service Sectors (ISS) and Leisure and Experience Venues (LEX) share an \"Aesthetics-centered driving\" mechanism, wherein Aesthetics (Aes) exerts a bidirectional impact on satisfaction. Higher aesthetic ratings significantly enhance satisfaction, whereas lower ratings cause a rapid decline. Notably, in the LEX category, Depression (Dpr) demonstrates a stronger negative impact on satisfaction than in ISS. The Retail and Trade Locations (RTL) sector presents an \"aesthetic paradox,\" where excessive aesthetics (Aes) may instead reduce satisfaction. Conversely, improvements in Vibrancy (Vib) substantially enhance satisfaction, underscoring the reliance of retail environments on stimulus-driven consumption.\u003c/p\u003e\u003cp\u003eSecond, the study clarifies the business-type heterogeneity of interaction effects among emotional dimensions. The interrelationships among emotional variables are both significant and divergent across categories. In CAT, Boring (Bro) and Wealth (Wth) exhibit strong interactive effects that modulate customer satisfaction. In ISS, the interaction between Aesthetics (Aes) and Wealth (Wth) enhances perceived service credibility through dual \"aesthetic\u0026ndash;capital\" cues. In LEX, the nonlinear interaction between Aesthetics (Aes) and Vibrancy (Vib) determines the maintenance of immersive experience. In RTL, a strong negative interaction between Aesthetics (Aes) and Depression (Dpr) reveals a \"cognitive contrast amplification\" effect, in which high aesthetic refinement under high-pressure conditions intensifies perceived spatial oppression.\u003c/p\u003e\u003cp\u003eTheoretically, this study advances the conceptual boundaries of the service atmosphere theory and emotional geography in three ways. (1) The outdoor extension of service atmosphere theory. Classical models emphasize the controllability of enclosed commercial spaces, whereas this study demonstrates that the streetscape itself functions as an effective service scape. Satisfaction is influenced not only by safety (Saf) perception but also by social and informal interactional cues, validating the notion that nonformal interactions (e.g., vendors, street conversations) contribute substantially to emotional generation. (2) The operational verification of emotional geography across business contexts. The study quantitatively confirms that the \"environmental emotion \u0026rarr; place satisfaction\" pathway varies across business types, indicating that the mechanism of emotional perception is jointly moderated by spatial functionality and consumption attributes. (3) The correction of linear assumptions and the introduction of nonlinear modeling. Using Random Forest and SHAP algorithms, the results reveal widespread threshold and nonlinear relationships between variables such as Safety (Saf), Aesthetics (Aes), and Vibrancy (Vib) and satisfaction, thereby overcoming the limitations of traditional linear models.\u003c/p\u003e\u003cp\u003eIn summary, supported by interpretable machine learning modeling, this study establishes a comprehensive analytical framework linking streetscape emotional perception to commercial satisfaction, unveiling the complex mechanisms through which urban streets generate affective and experiential responses. The findings provide data-driven and fine-grained decision-making references for street-space design and renewal, while offering practical and theoretical support for future exploration of \"data-driven emotional design\" and \"business-type-customized spatial interventions.\"\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThis work was supported by the National Social Science Foundation Arts Project of China [Grant No. 22CG182] and the National Natural Science Foundation of China [Grant No. 52208088].\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eYiwei Mo was responsible for the conceptualization and overall design of the study, performed data curation and management, conducted formal analysis, and drafted the original manuscript. Chao Luo and Jingjing Wang contributed to the writing of the original draft, was responsible for methodology design and data management, and secured project funding. Youpeng Lu contributed to the review and editing of the manuscript. Yunzhong Wang was in charge of project administration and coordination and participated in the review and editing of the manuscript. Qi Wu was responsible for visualization and assisted with formal analysis.Zixiang Feng was responsible for visualization and assisted with formal analysis.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eData will be made available on request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAshkezari-Toussi S, Kamel M, Sadoghi-Yazdi H (2019) Emotional maps based on social networks data to analyze cities emotional structure and measure their emotional similarity. Cities 86:113\u0026ndash;124. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cities.2018.09.009\u003c/span\u003e\u003cspan address=\"10.1016/j.cities.2018.09.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBitner MJ (1992) Servicescapes: The Impact of Physical Surroundings on Customers and Employees. J Mark 56(2):57\u0026ndash;71. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/002224299205600205\u003c/span\u003e\u003cspan address=\"10.1177/002224299205600205\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChawla NV, Bowyer KW, Hall LO, Kegelmeyer WP (2002) SMOTE: synthetic minority over-sampling technique. J Artif Intell Res 16:321\u0026ndash;357\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen S, Yu B, Shi G, Cai Y, Wang Y, He P (2025) Scale-Dependent Relationships Between Urban Morphology and Noise Perception: A Multi-Scale Spatiotemporal Analysis in New York City. Land 14(3). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/land14030476\u003c/span\u003e\u003cspan address=\"10.3390/land14030476\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen T-L, Chiu H-W, Lin Y-F (2020) How do East and Southeast Asian Cities Differ from Western Cities? A Systematic Review of the Urban Form Characteristics. Sustainability 12(6). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/su12062423\u003c/span\u003e\u003cspan address=\"10.3390/su12062423\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCorreia A, Kozak M (2016) Tourists' shopping experiences at street markets: Cross-country research. Tour Manag 56:85\u0026ndash;95. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.tourman.2016.03.026\u003c/span\u003e\u003cspan address=\"10.1016/j.tourman.2016.03.026\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCuesta-Mosquera AP, Wahl M, Acosta-L\u0026oacute;pez JG, Garc\u0026iacute;a-Reynoso JA, Aristiz\u0026aacute;bal-Zuluaga BH (2020) Mixing layer height and slope wind oscillation: Factors that control ambient air SO2 in a tropical mountain city. Sustainable Cities Soc 52. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.scs.2019.101852\u003c/span\u003e\u003cspan address=\"10.1016/j.scs.2019.101852\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDespotovic M, Hauser C (2025) A beautiful place: investigating the determinants of perceived scenic beauty in Austrian landscapes with social media data. Humanit Social Sci Commun 12(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1057/s41599-024-04317-2\u003c/span\u003e\u003cspan address=\"10.1057/s41599-024-04317-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDhakal CK, Khadka S (2021) Heterogeneities in Consumer Diet Quality and Health Outcomes of Consumers by Store Choice and Income. Nutrients 13(4). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/nu13041046\u003c/span\u003e\u003cspan address=\"10.3390/nu13041046\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDom\u0026iacute;nguez-Quintero AM, Gonz\u0026aacute;lez-Rodr\u0026iacute;guez MR, Rold\u0026aacute;n JL (2021) The role of authenticity, experience quality, emotions, and satisfaction in a cultural heritage destination. In \u003cem\u003eAuthenticity and Authentication of Heritage\u003c/em\u003e (pp. 103\u0026ndash;117). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4324/9781003130253-9\u003c/span\u003e\u003cspan address=\"10.4324/9781003130253-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDu F, Wang J, Mao L, Kang J (2024) Daily rhythm of urban space usage: insights from the nexus of urban functions and human mobility. Humanit Social Sci Commun 11(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1057/s41599-023-02577-y\u003c/span\u003e\u003cspan address=\"10.1057/s41599-023-02577-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDubey A, Naik N, Parikh D, Raskar R, Hidalgo CA (2016) Deep learning the city: Quantifying urban perception at a global scale. Lecture Notes in Computer Science\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEwing R, Handy S, Brownson RC, Clemente O, Winston E (2006) Identifying and Measuring Urban Design Qualities Related to Walkability. J Phys Activity Health 3(S1):S223\u0026ndash;S240. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1123/jpah.3.s1.s223\u003c/span\u003e\u003cspan address=\"10.1123/jpah.3.s1.s223\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFeizizadeh B, Omarzadeh D, Blaschke T (2024) Spatiotemporal mapping of urban trade and shopping patterns: A geospatial big data approach. Int J Appl Earth Obs Geoinf 128. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jag.2024.103764\u003c/span\u003e\u003cspan address=\"10.1016/j.jag.2024.103764\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eForoudi P, Cuomo MT, Foroudi MM, Katsikeas CS, Gupta S (2020) Linking identity and heritage with image and a reputation for competition. J Bus Res 113:317\u0026ndash;325\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGao S, Janowicz K, Couclelis H (2017) Extracting urban functional regions from points of interest and human activities on location-based social networks. Trans GIS 21(3):446\u0026ndash;467. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/tgis.12289\u003c/span\u003e\u003cspan address=\"10.1111/tgis.12289\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGao Y, Zhao J, Han L (2023) Quantifying the nonlinear relationship between block morphology and the surrounding thermal environment using random forest method. Sustainable Cities Soc 91. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.scs.2023.104443\u003c/span\u003e\u003cspan address=\"10.1016/j.scs.2023.104443\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGehl J (2011) Life between buildings\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHan C, Lieu SJ, Hwang U, Guhathakurta S (2025) Do streetscapes still matter for customer ratings of eating and drinking establishments in car-dependent cities? J Urban Des 1\u0026ndash;22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/13574809.2025.2541953\u003c/span\u003e\u003cspan address=\"10.1080/13574809.2025.2541953\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHe H, Garcia EA (2009) Learning from imbalanced data. IEEE Trans Knowl Data Eng 21(9):1263\u0026ndash;1284\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHe K, Zhang X, Ren S, Sun J (2016) \u003cem\u003eDeep Residual Learning for Image Recognition\u003c/em\u003e Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHo TP, Stevenson M, Thompson J, Nguyen TQ (2021) Evaluation of Urban Design Qualities across Five Urban Typologies in Hanoi. Urban Sci 5(4). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/urbansci5040076\u003c/span\u003e\u003cspan address=\"10.3390/urbansci5040076\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang Z, Ling X, Wang P, Zhang F, Mao Y, Lin T, Wang F-Y (2018) Modeling real-time human mobility based on mobile phone and transportation data fusion. Transp Res Part C: Emerg Technol 96:251\u0026ndash;269. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.trc.2018.09.016\u003c/span\u003e\u003cspan address=\"10.1016/j.trc.2018.09.016\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIstrate A-L (2025) Street vitality: what predicts pedestrian flows and stationary activities on predominantly residential Chinese streets, at the mesoscale? J Plann Educ Res 45(1):66\u0026ndash;80\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJacobs J (1961) The Death and Life of Great American Cities. Random House\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJang JY, Baek EJ, Yoon SY, Choo HJ (2018) Store Design: Visual Complexity and Consumer Responses. Int J Des 12(2):105\u0026ndash;118. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ijdesign.org/index.php/IJDesign/article/view/2934\u003c/span\u003e\u003cspan address=\"https://www.ijdesign.org/index.php/IJDesign/article/view/2934\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJani D, Han H (2011) Investigating the key factors affecting behavioral intentions. Int J Contemp Hospitality Manage 23(7):1000\u0026ndash;1018. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1108/09596111111167579\u003c/span\u003e\u003cspan address=\"10.1108/09596111111167579\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJaputra A, Ekinci Y, Simkin L (2019) Self-congruence, brand attachment and compulsive buying. J Bus Res 99:456\u0026ndash;463\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKleeman A, Giles-Corti B, Gunn L, Hooper P, Foster S (2023) The impact of the design and quality of communal areas in apartment buildings on residents' neighbouring and loneliness. Cities 133. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cities.2022.104126\u003c/span\u003e\u003cspan address=\"10.1016/j.cities.2022.104126\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKoo BW, Hwang U, Guhathakurta S (2023) Streetscapes as part of servicescapes: Can walkable streetscapes make local businesses more attractive? Comput Environ Urban Syst 106:102030\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKrizhevsky A, Sutskever I, Hinton GE (2012) ImageNet Classification with Deep Convolutional Neural Networks. Advances in neural information processing systems\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi X, Chen M, Wang R (2024) Assessing the nonlinear impact of green space exposure on psychological stress perception using machine learning and street view images. Front Public Health 12:1402536. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpubh.2024.1402536\u003c/span\u003e\u003cspan address=\"10.3389/fpubh.2024.1402536\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu K, Yin L, Lu F, Mou N (2020) Visualizing and exploring POI configurations of urban regions on POI-type semantic space. Cities 99. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cities.2020.102610\u003c/span\u003e\u003cspan address=\"10.1016/j.cities.2020.102610\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu Y, Chen M, Wang M, Huang J, Thomas F, Rahimi K, Mamouei M (2023) An interpretable machine learning framework for measuring urban perceptions from panoramic street view images. iScience 26(3):106132. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.isci.2023.106132\u003c/span\u003e\u003cspan address=\"10.1016/j.isci.2023.106132\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLundberg SM, Lee S-I (2017) A unified approach to interpreting model predictions. \u003cem\u003eAdvances in neural information processing systems\u003c/em\u003e, \u003cem\u003e30\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLuo P, Yu B, Li P, Liang P (2022) Spatially varying impacts of the built environment on physical activity from a human-scale view: Using street view data. Front Environ Sci 10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fenvs.2022.1021081\u003c/span\u003e\u003cspan address=\"10.3389/fenvs.2022.1021081\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLuo W, Wang F (2003) Measures of Spatial Accessibility to Healthcare in a GIS Environment: Synthesis and a Case Study in Chicago Region. Environ Plann B Plann Des 30(6):865\u0026ndash;884. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1068/b29120\u003c/span\u003e\u003cspan address=\"10.1068/b29120\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLynch K (1960) The Image of the City. MIT Press\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMa Z (2023) Deep exploration of street view features for identifying urban vitality: A case study of Qingdao city. Int J Appl Earth Obs Geoinf 123. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jag.2023.103476\u003c/span\u003e\u003cspan address=\"10.1016/j.jag.2023.103476\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMcKenzie G, Romm D (2021) Measuring urban regional similarity through mobility signatures. Comput Environ Urban Syst 89. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compenvurbsys.2021.101684\u003c/span\u003e\u003cspan address=\"10.1016/j.compenvurbsys.2021.101684\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMehrabian A, Russell JA (1974) An Approach to Environmental Psychology. MIT Press\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMehta V (2013) The street: a quintessential social public space. Routledge\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMeng Q, Sun Y, Kang J (2017) Effect of temporary open-air markets on the sound environment and acoustic perception based on the crowd density characteristics. Sci Total Environ 601\u0026ndash;602:1488\u0026ndash;1495. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.scitotenv.2017.06.017\u003c/span\u003e\u003cspan address=\"10.1016/j.scitotenv.2017.06.017\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMittal V, Kamakura WA (2009) The service quality\u0026ndash;satisfaction link revisited: exploring asymmetries and dynamics. J Acad Mark Sci 37(3):378\u0026ndash;390. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11747-009-0152-2\u003c/span\u003e\u003cspan address=\"10.1007/s11747-009-0152-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNaik N, Philipoom J, Raskar R, Hidalgo C (2014) \u003cem\u003eStreetscore-predicting the perceived safety of one million streetscapes\u003c/em\u003e Proceedings of the IEEE conference on computer vision and pattern recognition workshops\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePsyllidis A, Gao S, Hu Y, Kim EK, McKenzie G, Purves R, Yuan M, Andris C (2022) Points of Interest (POI): a commentary on the state of the art, challenges, and prospects for the future. Comput Urban Sci 2(1):20. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s43762-022-00047-w\u003c/span\u003e\u003cspan address=\"10.1007/s43762-022-00047-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eQin L, Sun J, Niu Q, Yuan M (2025) Residential spatial differentiation and influencing factors of permanent and temporary populations based on mobile signaling data: A case study of Wuhan, China. Appl Geogr 184. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.apgeog.2025.103760\u003c/span\u003e\u003cspan address=\"10.1016/j.apgeog.2025.103760\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eQin X, Zhen F, Gong Y (2019) Combination of Big and Small Data: Empirical Study on the Distribution and Factors of Catering Space Popularity in Nanjing, China. J Urban Plan Dev 145(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1061/(asce)up.1943-5444.0000489\u003c/span\u003e\u003cspan address=\"10.1061/(asce)up.1943-5444.0000489\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eQiu Y, Wu M, Huang Q, Kang Y (2025) Do You Know Your Neighborhood? Integrating Street View Images and Multi-task Learning for Fine-Grained Multi-Class Neighborhood Wealthiness Perception Prediction. \u003cem\u003eCities\u003c/em\u003e, \u003cem\u003e158\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cities.2025.105703\u003c/span\u003e\u003cspan address=\"10.1016/j.cities.2025.105703\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eReber R, Schwarz N, Winkielman P (2004) Processing fluency and aesthetic pleasure: Is beauty in the perceiver\u0026rsquo;s processing experience? Personality Social Psychol Rev 8(4):364\u0026ndash;382. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1207/s15327957pspr0804_3\u003c/span\u003e\u003cspan address=\"10.1207/s15327957pspr0804_3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSalesses MP (2012) Place Pulse: Measuring the collaborative image of the city. Massachusetts Institute of Technology]\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShukri SM, Wahab MH, Awaluddin ZL, Aminuddin AMR, Hasan MI (2022) The Role of Attachment in Creating Sustainable Sense of Place for Traditional Streets in Alor Setar, Malaysia. J Des Built Environ 22(1):55\u0026ndash;71\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSimonyan K, Zisserman A (2015) Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv preprint. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003eIssue. https://arxiv.org/abs/1409.1556\u003c/span\u003e\u003cspan address=\"http://Issue. https://arxiv.org/abs/1409.1556\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStandardization Administration of, C., \u0026amp; National Bureau of Statistics of, C (2017) GB/T 4754\u0026ndash;2017 国民经济行业分\u0026amp;#31867\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStroebele N, De Castro JM (2004) Effect of ambience on food intake and food choice. Nutrition 20(9):821\u0026ndash;838. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.nut.2004.05.012\u003c/span\u003e\u003cspan address=\"10.1016/j.nut.2004.05.012\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003evan Veghel J, Dane G, Agugiaro G, Borgers A (2024) Human-centric computational urban design: optimizing high-density urban areas to enhance human subjective well-being. Comput Urban Sci 4(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s43762-024-00124-2\u003c/span\u003e\u003cspan address=\"10.1007/s43762-024-00124-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVerhoef PC, Lemon KN, Parasuraman A, Roggeveen A, Tsiros M, Schlesinger LA (2009) Customer Experience Creation: Determinants, Dynamics and Management Strategies. J Retail 85(1):31\u0026ndash;41. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jretai.2008.11.001\u003c/span\u003e\u003cspan address=\"10.1016/j.jretai.2008.11.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWalsh G, Shiu E, Hassan LM, Michaelidou N, Beatty SE (2011) Emotions, store-environmental cues, store-choice criteria, and marketing outcomes. J Bus Res 64(7):737\u0026ndash;744. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jbusres.2010.07.008\u003c/span\u003e\u003cspan address=\"10.1016/j.jbusres.2010.07.008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang J, Gao C, Wang M, Zhang Y (2023) Identification of Urban Functional Areas and Urban Spatial Structure Analysis by Fusing Multi-Source Data Features: A Case Study of Zhengzhou, China. Sustainability 15(8). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/su15086505\u003c/span\u003e\u003cspan address=\"10.3390/su15086505\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang Y, Chen X, Gao M, Dong J (2022) The use of random forest to identify climate and human interference on vegetation coverage changes in southwest China. Ecol Ind 144. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ecolind.2022.109463\u003c/span\u003e\u003cspan address=\"10.1016/j.ecolind.2022.109463\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWei J, Yue W, Li M, Gao J (2022) Mapping human perception of urban landscape from street-view images: A deep-learning approach. Int J Appl Earth Obs Geoinf 112:102886\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWhyte WH (1980) \u003cem\u003eThe Social Life of Small Urban Spaces. Conservation Foundation\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu F, Zhen F, Qin X, Wang X, Wang F (2018) From central place to central flow theory: an exploration of urban catering. Tourism Geographies 21(1):121\u0026ndash;142. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/14616688.2018.1457076\u003c/span\u003e\u003cspan address=\"10.1080/14616688.2018.1457076\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang C, Zhang Y (2024) Public emotions and visual perception of the East Coast Park in Singapore: A deep learning method using social media data. Urban Forestry Urban Green 94. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ufug.2024.128285\u003c/span\u003e\u003cspan address=\"10.1016/j.ufug.2024.128285\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang R, Zhang J, Xu Q, Luo X (2020) Urban-rural spatial transformation process and influences from the perspective of land use: A case study of the Pearl River Delta Region. Habitat Int 104. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.habitatint.2020.102234\u003c/span\u003e\u003cspan address=\"10.1016/j.habitatint.2020.102234\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYi Y-M, Gim T-H (2018) What Makes an Old Market Sustainable? An Empirical Analysis on the Economic and Leisure Performances of Traditional Retail Markets in Seoul. Sustainability 10(6). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/su10061779\u003c/span\u003e\u003cspan address=\"10.3390/su10061779\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eY\u0026uuml;ksel A, Y\u0026uuml;ksel F, Bilim Y (2010) Destination attachment: Effects on customer satisfaction and cognitive, affective and conative loyalty. Tour Manag 31(2):274\u0026ndash;284. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.tourman.2009.03.007\u003c/span\u003e\u003cspan address=\"10.1016/j.tourman.2009.03.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang D, Chen M, Huang W, Gong Y, Zhao K (2024) Exploring urban semantics: A multimodal model for POI semantic annotation with street view images and place names. Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang E, Hou J, Long Y (2025) The form of China\u0026rsquo;s urban commercial expansion in the digital era. Nat Cities 2(7):639\u0026ndash;649. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s44284-025-00254-6\u003c/span\u003e\u003cspan address=\"10.1038/s44284-025-00254-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang E, Xie H, Long Y (2023) Decoding the association between urban streetscape skeletons and urban activities: Experiments in Beijing using Dazhong Dianping data. Trans Urban Data Sci Technol 2(1):3\u0026ndash;18\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou H, Gu J, Liu Y, Wang X (2022) The impact of the skeleton and skin for the streetscape on the walking behavior in 3D vertical cities. Landsc Urban Plann 227:104543\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou X, Noulas A, Mascolo C, Zhao Z (2018) \u003cem\u003eDiscovering Latent Patterns of Urban Cultural Interactions in WeChat for Modern City Planning\u003c/em\u003e Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery \u0026amp; Data Mining\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhu D, Song D, Zhu B, Zhao J, Li Y, Zhang C, Zhu D, Yu C, Han T (2024) Understanding complex interactions between neighborhood environment and personal perception in affecting walking behavior of older adults: A random forest approach combined with human-machine adversarial framework. Cities 146. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cities.2023.104737\u003c/span\u003e\u003cspan address=\"10.1016/j.cities.2023.104737\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Street-level Emotion Perception, Service Landscape Theory, Commercial Satisfaction, Business Format Heterogeneity, Random Forest, Emotion-Customized Spatial Intervention","lastPublishedDoi":"10.21203/rs.3.rs-8017731/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8017731/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIn cross-disciplinary research on urban planning and service marketing, the street landscape is seen as an extended service environment. However, most studies focus on the link between a single business type and one emotional dimension. This leaves the complex mechanisms between streetscape emotional perception and commercial satisfaction unrevealed. This study combines street-view imagery, Dianping point-of-interest (POI) data, and online ratings to build a dataset. It covers four main commercial categories (Catering Services, Retail and Trade Locations (RTL), In-store Service Sectors, Leisure and Experience Venues) and six emotional perceptions (Safety, Aesthetics, Wealth, Vibrancy, Depression, Boring). The study uses Random Forest (RF) regression and SHapley Additive exPlanations (SHAP) to examine how streetscape emotions influence businesses. Results show catering services are mostly affected by the Depression and Vibrancy (Dpr\u0026ndash;Vib) link. In-store services and leisure venues are shaped by the synergy of Aesthetics and Wealth (Aes\u0026ndash;Wth). Retail trade reveals a tension between Aesthetics and Depression. The study proposes a business-type-specific \"emotion-customized\" spatial intervention framework to guide urban renewal and management of mixed-use streets.\u003c/p\u003e","manuscriptTitle":"The Influence of Street-Scene Emotional Perception on Commercial Satisfaction: A Study Based on Random Forest Model and SHAP Algorithm","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-01 08:56:10","doi":"10.21203/rs.3.rs-8017731/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-02-28T16:47:52+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-01-01T11:09:39+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-17T14:42:40+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"44568926178085901768125499634916887924","date":"2025-12-08T12:14:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"217871180038179161578770408514245163777","date":"2025-12-08T08:31:55+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-11-26T11:34:57+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-11-12T06:45:09+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-11-11T05:17:26+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-11-04T09:48:12+00:00","index":"","fulltext":""},{"type":"submitted","content":"Humanities and Social Sciences Communications","date":"2025-11-03T09:54:04+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"15c582bc-4924-421f-9c90-aa7fd809ea19","owner":[],"postedDate":"December 1st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":58773909,"name":"Earth and environmental sciences/Environmental social sciences"},{"id":58773910,"name":"Social science/Environmental studies"},{"id":58773911,"name":"Scientific community and society/Geography"},{"id":58773912,"name":"Social science/Geography"}],"tags":[],"updatedAt":"2026-03-27T02:23:14+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-01 08:56:10","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8017731","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8017731","identity":"rs-8017731","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}