Research on Quality and Safety Risk Identification of Import and Export Toys Based on the WOA-BP Model

Research on Quality and Safety Risk Identification of Import and Export Toys Based on the WOA-BP Model | 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 Research on Quality and Safety Risk Identification of Import and Export Toys Based on the WOA-BP Model Qiong He, Pengzhao Li This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6416394/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 13 You are reading this latest preprint version Abstract Within the scope of customs risk management, the weakness in the ability to identify quality and safety risks of import and export products is a key factor contributing to the persistently high product recall rate. To effectively improve this situation, this paper selects import and export toy products as research samples and constructs a research system framework of "data collection - risk classification - risk identification". This study establishes a quality and safety risk identification model for import and export toys based on two machine learning techniques, namely Latent Dirichlet Allocation (LDA) and neural networks. Firstly, Use Python to preprocess the information and employ the ROSTCM software to conduct word frequency analysis to obtain a network relationship diagram. Based on the Dirichlet distribution topic model and toy safety-related indicators, keywords are extracted to determine toy safety risk factors as well as toy safety risk events. A safety risk identification model for import and export toys is established through the BP neural network, and the Whale Optimization Algorithm (WOA) is used to optimize the model. The results of the simulation study show that in terms of the model's accuracy, the prediction accuracy rate of the WOA-BP model is 95.71%, which is 5.71 percentage points higher than that of the BP model. In terms of the model's regression performance, the R-value of the WOA-BP model's test set is 0.997. The WOA-BP model is superior to the BP model and its prediction results are more in line with the actual situation, enabling it to better accomplish the task of identifying quality and safety risks. Earth and environmental sciences/Environmental social sciences/Sustainability Physical sciences/Mathematics and computing/Computer science Physical sciences/Mathematics and computing/Information technology Physical sciences/Mathematics and computing/Statistics Import and Export Toys WOA-BP Neural Network Risk Identification Topic Model Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction In the field of international trade, risk management of the quality and safety of imported and exported products is a crucial link in ensuring the continuous, stable, and healthy development of trade activities[ 1 ]. Currently, different countries vary in their capabilities regarding the risk management of the quality and safety of imported and exported products. Among them, the US Customs is proficient in integrating resources in risk management and focuses on establishing an intelligence collection system, having developed automated identification systems such as ATS-G and ACE[ 2 ]. France, on the other hand, conducts risk identification through three levels, namely the national, EU, and international levels, and uses big data algorithm technology to optimize the data pool[ 3 ]. China's Customs combines import and export data and periodically selects cross-border e-commerce imported consumer goods that draw high consumer attention, have high quality and safety risks, and experience high growth rates in imports for quality inspections. Although different countries all have a complete set of approaches for the risk management of the quality and safety of imported and exported products, they are all constructed based on existing models and technological frameworks. When dealing with the increasingly complex and changeable international trade environment, there will be limitations, and it is difficult to accurately and efficiently identify all potential risks[ 4 ]. Stable and efficient risk identification technology is the foundation of risk management and can provide relevant standards for subsequent risk assessment and response. Toys, as special and sensitive products, are different from other imported and exported products, featuring compliance with diverse standards and regulations, as well as high-frequency contact and repeated use[ 5 ]. Therefore, this study selects imported and exported toys as the research object, constructs a set of quality and safety risk identification models for imports and exports, analyzes the risk factors of imported and exported toys, and identifies their quality and safety risks, with the aim of providing a novel method for the identification of quality and safety risks of imported and exported products. Risk identification technology is a key link in the quality and safety risk management of import and export toys by customs. With the advent of the big data era, risk identification technology has also been continuously advancing. Currently, domestic and foreign scholars have applied neural network and other technologies in the field of risk identification. Guozhu Cheng constructed a dangerous behavior spectrum to classify the risk levels of dangerous driving behaviors and used the SVM statistical learning method to complete driving risk identification[ 6 ]. Aakanshi Gupta analyzed 8 machine learning algorithms such as support vector machines, random forests, and neural networks, collected relevant clinical data and used 8 machine algorithms to identify stroke risks. The results showed that the neural network model had high accuracy and normalization and stood out among these 8 machine learning algorithms[ 7 ]. Alexandra Ovsyannikova constructed a feedforward neural network model and trained it using batch normalization, specific weight initialization, and random gradient descent methods to solve the problem of contract default risk identification[ 8 ]. Yanbo Jiang constructed a land use classification model based on the BP neural network and combined it with a hydrological model and ArcGIS spatial analysis tools to identify flood risk[ 9 ]. Although neural network technology has achieved many results in the field of risk identification, its application also has some disadvantages[ 10 ]. The currently popular BP neural network has problems such as slow learning speed, easy falling into local extrema, and uncertain values of network structure parameters when facing massive data. In response to this situation, Congfen Liu et al. used the SAE algorithm to optimize the BP neural network model to complete the research on audit risk identification, and the accuracy rate of the optimized model was greatly improved[ 11 ]. Xueting Zhao combined the principal component analysis method to optimize the BP neural network and improved the accuracy rate of the model for identifying vehicle operation risks in congested environments[ 12 ]. X.L. Ji combined the NMM algorithm with the WOA-BP neural network to complete the identification of hole defects in heat conduction problems by detecting circular and elliptical hole defects[ 13 ]. In addition, Yali Li et al. found that among several new intelligent optimization algorithms, the WOA algorithm has stronger adaptability, faster convergence speed, and simple parameter settings compared with other algorithms[ 14 ]. When facing import and export commodity information with a large data scale, the WOA algorithm can identify potential quality and safety risk factors more quickly and comprehensively. In view of this, this study intends to use the WOA algorithm to optimize the BP neural network to solve problems such as the easy falling into local optimum and prediction instability of a single BP neural network, thereby improving the global optimization ability. Neural networks have relatively strict requirements for input data. In the identification of import and export toy safety risks, toys come from a wide range of sources, and production standards vary in different countries and regions, resulting in diverse types of potential safety hazards in toys. The reasonable classification of toy information and an appropriate input set are of great significance for the accuracy of model results. Therefore, how to reasonably classify toy safety risk factors is the key to completing risk identification. In recent years, various fields have achieved rich results in risk factor classification. Wang et al. combined TF-IDF and Word2Vec and used the text clustering method to identify high-frequency and low-severity building safety report risks[ 15 ]. Weiwei Liu improved the LDA model by building a term dual optimization model, introduced the TF-IDF algorithm for text feature selection, optimized word weights using the Gaussian function, and obtained 26 risk topics for visualization by text clustering through the Sankey diagram[ 16 ]. Shangying Xu clustered risk topics of dual-source data sets based on the Sent-LDA model, identified 27 important risk points faced by equipment manufacturing enterprises, and performed discrete evolution analysis on the risk points[ 17 ]. Ruizhen Song et al. constructed a corpus, calculated the perplexity of text data, and identified railway engineering ecological risks based on the LDA "topic-word" distribution. Finally, 15 topic words with the largest correlation coefficients were obtained[ 18 ]. The LDA topic model can integrate and transform large-scale unquantifiable text data into structured data and extract valuable intelligence information. This study employs the LDA topic model to summarize the quality and safety text data of import and export toys into topics for classifying toy safety factors, facilitating the learning and training of the subsequent neural network model and enhancing the identification accuracy. The quality and safety risk identification of toys is the first step in risk management. Efficient and accurate risk identification can improve the risk management ability of import and export toy commodity quality and safety. In existing research, there is no objective and systematic identification of toy quality and safety risks. Based on this, this article takes the quality and safety of import and export toys as the research object and proposes a research framework of "data collection-risk classification-risk identification" based on two machine learning techniques: the topic model and the neural network. 1.1 Construction of the Import and Export Quality and Safety Risk Identification Model According to the recall historical text data of import and export toy commodities, based on two machine learning techniques: topic analysis and neural network, a framework of the quality and safety risk identification model for import and export toys is constructed, as shown in Fig. 1 . The model includes three parts: data collection, risk classification, and risk identification. In the data collection module, crawler technology is used to obtain recall information related to import and export toys from customs government websites. For the crawled text content, text cleaning, Chinese word segmentation, stop word removal, and text data initialization operations are carried out in sequence to complete text data preprocessing. The risk classification module includes two sub-modules: word frequency analysis and topic analysis. In the word frequency analysis sub-module, quality and safety keywords are first set, and then, based on ROSTCM software, keyword extraction is performed on the text information of import and export toys, TF-IDF word frequency statistics are carried out to form a correlation matrix, and the visualization processing of the word frequency matrix is completed. In the topic analysis sub-module, the LDA topic model is used to classify the screened keywords, screen out key topics and topic words, and divide toy safety risk factors and toy safety risk events based on this, providing data support for the subsequent model input and output environment. In the risk identification module, a quality and safety risk identification model for import and export toys is constructed based on the BP neural network model, and the WOA algorithm is used to optimize the parameters of the neural network to improve the accuracy of model identification. The model has three advantages. First, feature marking can simplify the workload. For a large amount of toy information, only the topic words after LDA topic analysis need to be feature marked to obtain the systematic import and export toy safety risk factors and toy safety risk events. Second, the word frequency analysis and topic analysis based on ROSTCM software and unsupervised machine learning models can avoid problems caused by the subjectivity of human clustering. Third, using the WOA algorithm to optimize the neural network model is expected to improve the overall identification accuracy of the model. The organization of this paper is summarized as follows: Section 2 conducts data processing and LDA topic classification; Section 3 constructs the neural network model; Section 4 analyzes and compares the results of the model; and Section 5 summarizes this paper and discusses the directions for future research. 2. Classification of Quality and Safety Risk Factors of Import and Export Toys Based on LDA Model 2.1 Data Collection This article uses relevant government websites and customs data as data sources and uses Python's data crawling technology to extract text data related to the recall of import and export toys. At the same time, this article extracts the EU RAPEX notification system, the US CPSC notification system, and the Canadian HC system from government websites and collects recall information about toys from 2010 to 2024. Among them, when screening the notification information of the US and Canada, keywords "toy" and "recall" are determined and screened in the title and abstract. Since the EU notification system reports toys and other commodities together, the request library is first used to obtain relevant reports in the web page, and then the content about toy recalls is extracted from the Beautifutlsup library to collect data from various countries. The text results are shown in Table 1 . Since the information contained in the title and keywords of toy recalls is limited, it may not be possible to obtain accurate toy risk types when used as data input into the topic model. Therefore, the toy defect content mentioned in the text is retained. 75% of the sample data is randomly selected from the database and divided into the training set, and the remaining 25% is used as the test set for subsequent model verification and analysis. Table 1 Data Collection and Preprocessing Results Database RAPEX CPCS HC Text Total Total Search Texts 3258 343 686 4287 Obtainable Texts 3141 145 440 3726 Texts after Deleting Duplicates 3130 138 374 3642 2.2 Word Frequency Analysis TF-IDF is a commonly used weighting technique for information retrieval and data mining, which can more accurately reflect the key degree of a word in a specific document and distinguish its importance weight[ 19 ]. ROSTCM is software mainly used for analyzing and processing text data. It uses natural language processing and text mining techniques to extract valuable information. In this study, based on ROSTCM software, TF-IDF word frequency analysis is performed on text data through functions such as dictionary setting, synonym setting, and part-of-speech screening. According to the analysis results, the top 30 high-frequency words ranked by TF-IDF values are listed, as shown in Table 2 . Table 2 High-Frequency Words in Import and Export Toy Recall Information Word TF-IDF Word TF-IDF Word TF-IDF Asphyxiation 0.056657 Detachable 0.028335 Strangulation 0.018003 Ingestion 0.053896 Soft toys 0.026622 Rubber toy 0.014665 Plastic 0.053784 Pollution 0.025059 Magnetic flux 0.010894 Reproductive 0.048236 Fall off 0.022302 Sound pressure 0.010764 Small parts 0.046861 Plush toys 0.020829 Electric toys 0.010677 Plastic dolls 0.041271 Hearing 0.019754 Rattle toys 0.010559 Phthalic acid 0.039328 Burns 0.020718 Flammable 0.009698 Toy clay 0.043076 Plastic toys 0.018515 Cuts 0.010238 Laceration 0.033570 Fiber materials 0.018077 Metal 0.008908 Components 0.031730 Toy sets 0.017260 Ileus 0.008290 The network relationship diagram can show the complex relationship structure and understand the interrelationships between various entities[ 20 ]. Based on the word frequency matrix obtained after software processing, which reflects the correlation strength between words through the co-occurrence frequency of words, the matrix is visualized to obtain the network relationship diagram of high-frequency words in import and export toy recall information, as shown in Fig. 2 . Combined with the network diagram, the relationship analysis of some high-frequency words was conducted. "Asphyxiation" and "Ingestion" are in the core positions in the associated network. They are closely linked with "Small parts", "Small components" and "Easy to fall off", which means that toy parts are prone to falling off, putting children at risk of asphyxiation or accidentally ingesting them. Words related to "Plastic" and its derivatives frequently appear. Among them, "Plastic" is associated with " Contains phthalates", and the latter is related to "Reproductive ", indicating that when plastic toys contain excessive amounts of this substance, it will endanger children's reproductive health. "Toy clay" is closely associated with "Environmental pollution", suggesting that toys may cause pollution to the environment during the use or disposal stages, resulting in their recall. Meanwhile, words like "Burns", "Cuts" and "Strangulation" are presented around "Personal injury", showing that the situations where toys endanger personal safety during the use process can also become the key factors for toy recalls. Through the above analysis, it can be found that there is a certain degree of correlation among different words. Although the frequency of occurrence of each word in the text and their mutual correlation have been clearly sorted out at the data level, the obtained output results are fragmented, and it is still unclear around which core themes the overall text unfolds. Therefore, it is necessary to classify the keys of the text and screen out the core themes and their corresponding theme words. 2.2 Word Frequency Analysis TF-IDF is a commonly used weighting technique for information retrieval and data mining, which can more accurately reflect the key degree of a word in a specific document and distinguish its importance weight[ 19 ]. ROSTCM is software mainly used for analyzing and processing text data. It uses natural language processing and text mining techniques to extract valuable information. In this study, based on ROSTCM software, TF-IDF word frequency analysis is performed on text data through functions such as dictionary setting, synonym setting, and part-of-speech screening. According to the analysis results, the top 30 high-frequency words ranked by TF-IDF values are listed, as shown in Table 2 . The network relationship diagram can show the complex relationship structure and understand the interrelationships between various entities[ 20 ]. Based on the word frequency matrix obtained after software processing, which reflects the correlation strength between words through the co-occurrence frequency of words, the matrix is visualized to obtain the network relationship diagram of high-frequency words in import and export toy recall information, as shown in Fig. 2 . Combined with the network diagram, the relationship analysis of some high-frequency words was conducted. "Asphyxiation" and "Ingestion" are in the core positions in the associated network. They are closely linked with "Small parts", "Small components" and "Easy to fall off", which means that toy parts are prone to falling off, putting children at risk of asphyxiation or accidentally ingesting them. Words related to "Plastic" and its derivatives frequently appear. Among them, "Plastic" is associated with " Contains phthalates", and the latter is related to "Reproductive ", indicating that when plastic toys contain excessive amounts of this substance, it will endanger children's reproductive health. "Toy clay" is closely associated with "Environmental pollution", suggesting that toys may cause pollution to the environment during the use or disposal stages, resulting in their recall. Meanwhile, words like "Burns", "Cuts" and "Strangulation" are presented around "Personal injury", showing that the situations where toys endanger personal safety during the use process can also become the key factors for toy recalls. Through the above analysis, it can be found that there is a certain degree of correlation among different words. Although the frequency of occurrence of each word in the text and their mutual correlation have been clearly sorted out at the data level, the obtained output results are fragmented, and it is still unclear around which core themes the overall text unfolds. Therefore, it is necessary to classify the keys of the text and screen out the core themes and their corresponding theme words. 2.3 Topic Analysis To discover the hidden features, relationships, and patterns in the data and make a reasonable classification of risk factors, this paper extracts the information contained in the high-frequency words and network relationships screened by word frequency analysis through topic analysis. The topic model based on LDA is adopted to screen out the risk factors caused by toy quality and safety[ 21 ]. LDA analysis is performed using the sklearn machine learning library in Python, and the optimal number of topics (K) is determined by perplexity. Perplexity is an effective method for evaluating and assisting in improving the parameters of a language probability model. By calculating the perplexity values corresponding to different numbers of topics and using the Python third-party toolkit Matplotlib to draw a graph, a line graph of the change in perplexity with the number of topics is obtained. When the perplexity is the lowest, the corresponding K value is the optimal. After analysis, when the K value is 4, the model's perplexity is at a minimum value, as shown in Fig. 3 . According to the perplexity calculation, the number of topics in this study is determined to be 4. Subsequently, representative topic words are selected from the top 10 keywords with the highest frequency in the 4 groups of topic words, and the topic is named according to the topic words. The LDA topic clustering model is used for clustering analysis, and the clustering results are shown in Fig. 4 . Figure 4 lists the top 30 most significant topic words related to the topic. According to the relevant standards of import and export toys, combined with the number of topics and keywords, they can be divided into toy types, toy materials, toy defects, and risk events. The specific topic words included in each topic are shown in Table 3 . Table 3 Topics and Corresponding Topic Words Divided by the LDA Topic Model Topic Topic Words Toy Types Plush Toys, Electric Toys, Handmade Toys, Soft Toys, Plastic Toys, Rattle Toys, Decompression Toys, Kitchen Toys, Bath Toys, Magnetic Toys, Fishing Toys Toy Materials Plastic, Wood, Plush, Metal, Paper, Rubber, Glass, Fiber Filling Materials Toy Defects Easily Fallen Off, Easily Detachable, Insufficient Stability, Easily Broken, Easily Fractured, Small Parts, Small Components, Sound Pressure Level, Flammable, Containing Excessive Phthalic Acid, High Magnetic Flux, Corrosion, Hygroscopicity, High Lead Concentration, High Cadmium Concentration, Sound Pressure Level, Dampness, Sharp Edges, Bacteria, Parasites, Warning Labels, Age Labels, Ingredient Labels, Instructions for Use, Safety Tips, Production Information Injury Events Reproductive System, Asphyxiation, Ingestion by Mistake, Personal Injury, Strangulation, Environmental Pollution, Hearing Impairment, Burns, Cuts, Intestinal Obstruction, Stabbing, Vision Impairment, Microbial Pollution, Chemical Hazards 2.4 Classification of Safety Risk Factors The LDA topic model, with its ability to deeply mine the semantic structure of text, extracts groups of words scattered throughout the text and related to different aspects of toys. Although it has initially outlined the contours of each topic, the relatively large number of topic words presents a certain degree of complexity and redundancy. In the subsequent neural network training process, this will lead to a high data dimension, triggering the "curse of dimensionality"[ 22 ]. Therefore, it is necessary to further summarize and integrate the topic words. Based on the topic words and combined with the basic technical requirements to be complied with in the "Toy Safety Basic Specification," the toy safety risk factors and toy safety risk events are classified. After topic analysis, a total of 14 toy safety risk events are screened out. Combined with the injuries caused by toy design defects, manufacturing processes, or materials listed in the specification, such as poisoning and other harmful substances injuries, asphyxiation, swallowing or inhaling foreign objects, electric shock, and other mechanical injuries including cuts, tears, abrasions, eye injuries, head injuries, and auditory injuries, these 14 safety risk events are classified and finally summarized into 5 categories: mechanical injuries, asphyxiation injuries, chemical injuries, hygienic injuries, and sensory injuries. Table 4 Selection of Risk Feature Markers Risk Factor Specific Description Feature Marker Toy Types Plastic Toys, Electric Toys, Metal Toys, Doll Toys, Other Toys A Toy Materials Plastic Materials, Metal Materials, Textile Materials, Composite Materials, Other Materials B Toy Defects Physical Component Defects, Material Property Defects, Chemical Substance Defects, Biological Pollution Defects, Labeling Defects, Other Factors C After topic analysis, a total of 26 toy defects are screened out. Based on the mechanical physical safety testing standard system, the combustion safety testing standard system, the chemical element migration safety standard system, and the electromagnetic compatibility performance testing standard system in the toy safety testing standard system[ 23 ], the toy defects are divided according to the physical, chemical, biological, and labeling defects of the toy materials. Among them, factors with a word frequency of less than 5 times are classified as other factors. The factors that also affect toy risk identification include toy types and the materials of the toys themselves. Different toy types have differences in applicable ages and functional characteristics. After topic analysis, a total of 11 toy types are screened out. Based on the 3C toy certification classification in China, the toy types are divided into 5 categories, and factors with a word frequency of less than 5 times are classified as other factors. Different materials of toys will cause different safety risks. After topic analysis, a total of 8 toy materials are screened out and summarized into plastic types; metal types; rubber types; textile types; other materials. Feature markers are assigned to different risk factors. Toy types are marked as feature A; the material properties of the toys themselves are marked as feature B; and toy defects are marked as feature C, as shown in Table 4 . Through further division and integration of the topic words, redundant and interfering factors are eliminated to improve the accuracy and stability of the model in the prediction process. The feature-marked items are used as toy risk factors and applied to the input environment of the subsequent neural network, and the toy safety risk events are applied to the output environment of the subsequent neural network. 3. Construction of the Quality and Safety Risk Identification Model for Import and Export Toys 3.1 Construction of the BP Neural Network Model This study uses MATLAB software programming to implement the construction of the BP neural network model. The BP neural network model is usually composed of an input layer, a hidden layer, and an output layer. The features A, B, and C are used as the data source for the input layer of the BP neural network, and the number of input layer nodes is set to 3. The number of hidden layer nodes in the BP neural network has a great impact on the convergence speed and prediction accuracy of the network. If the number of nodes is too small, the network training error will be large, the iteration time will be long, and the identification accuracy of the predicted samples will be low. If the number of nodes is too large, the training time will be extended, and the training network will be overfitted[ 24 ]. Therefore, determining an appropriate number of hidden layer nodes is extremely critical. In this study, the number of hidden layer nodes is determined according to the formula (1): \(\:M=\sqrt{n+m}+c\) (1) In formula (1), \(\:n\) is the number of input nodes; \(\:m\) is the number of output nodes; \(\:c\) is a constant from 1 to 10; \(\:M\) is the number of hidden layer nodes. According to the formula, the number of hidden layers in this study is determined to be 12. The output layer is set with 5 nodes, and each node corresponds to a category of toy risk events. Since there are 5 types of toy risk events, namely mechanical injuries, asphyxiation injuries, chemical injuries, hygienic injuries, and sensory injuries, during the training stage, the category labels are converted into one-hot encoded vector forms suitable for processing by the output layer of the neural network using the ind2vec function. After the prediction results are denormalized, the vector form output by the neural network is converted back into category labels using the vec2ind function to achieve correspondence and comparison with the original category labels. In addition, since each node in the input layer of this study contains different classifications and the parameter quantities of each node differ greatly, it is difficult to adjust the threshold values between layers, affecting the convergence rate and network accuracy. Therefore, the input parameters are normalized according to formula (2): \(\:y=\frac{(x-min(X\left)\right)\times\:(ne{w}_{max}-ne{w}_{min})}{\left(max\right(X)-min(X\left)\right)}+ne{w}_{min}\) (2) In formula (2), \(\:X\) is the original data vector, \(\:min\left(X\right)\) is the minimum value in \(\:X\) , \(\:max\left(X\right)\) is the maximum value in \(\:X\) , \(\:ne{w}_{min}\) and \(\:ne{w}_{max}\) are the minimum and maximum values of the specified normalization interval (usually − 1 and 1). By uniformly normalizing the input parameters, the impact of data distribution differences on the generalization ability of the model can be reduced, so that the laws learned by the model on the training set can be better applied to the test set and practical applications. 3.2 Construction of the WOA-BP Network Model The WOA algorithm simulates the group hunting behavior of whales and searches for the optimal solution through the update and search strategies of the whale's position[ 25 ]. By combining the WOA algorithm with the BP neural network and using the global search ability of the WOA algorithm to optimize the weights and thresholds of the BP neural network, the performance of the BP neural network can be improved 26. The WOA algorithm calculates the fitness based on toy safety data. By applying the BP neural network under different parameter combinations to predict risk data and determining the fitness according to the degree of matching with the actual situation and comparing with the global optimal fitness after each calculation to update the global optimal position, which represents the current optimal parameter combination. In the WOA optimization algorithm, the random number p first divides the position update strategy of the search agent into individual selection-based or spiral upward. |A| further determines whether to approach the global optimal individual or a randomly selected individual under the selected type. When p < 0.5 and |A| < 1, the search agent approaches the global optimal individual. If |A| ≥ 1, the search agent approaches a randomly selected individual. When p ≥ 0.5, the search agent adopts a spiral update strategy. After each update, the fitness is recalculated, and the individual and global optimal positions are updated to promote the algorithm to evolve towards the optimal parameter combination. When the maximum iteration number is reached, the optimal parameter combination can be output, and then the training and prediction of the toy quality and safety risk identification model can be carried out. The optimization process of the WOA algorithm is shown in Fig. 5 . This study defines a target function to measure the quality of each whale individual's position in the algorithm. When setting the target function, considering that the example is a multi-classification task, the way of minimizing the root mean square error is used for setting. The initial population size is 20, the maximum iteration number is 50, and the optimization parameter range is set between − 1 and 1. 4. Results Analysis To comprehensively evaluate the level of the WOA-BP model in identifying the quality and safety risks of import and export toys, the unoptimized BP model and the WOA-BP model are simulated and tested on the MATLAB 2022 platform. The ability of the WOA-BP model is judged through the accuracy analysis and regression analysis of the models. 4.1 Accuracy Performance Analysis The prediction accuracy results of the BP model and the WOA-BP model for toy risk events are shown in Fig. 6 . It can be seen from the figure that in the test set, the prediction accuracy rate of the BP model is 90%, while the prediction accuracy rate of the WOA-BP model is 95.71%. The WOA-BP model has higher accuracy and smaller errors. Therefore, the prediction accuracy of the WOA-BP model is much better than that of the BP model, indicating that the WOA-BP model can better learn the characteristics and laws of the data and has good generalization ability. 4.2 Regression Analysis The regression abilities of the BP model and the WOA-BP model are shown in Fig. 7 . The R value of the BP model's test set is 0.938, and the R value of the WOA-BP model's test set reaches 0.997. It can be seen from the figure that the regression effect of the WOA-BP test is better, and the fitting degree with the target value is higher. Overall, the WOA-BP test performs better than the BP test in this regression analysis. In summary, the WOA-BP model performs better than the BP model in predicting toy risk events. The WOA-BP model is more excellent in terms of accuracy and regression ability, has a higher fitting degree with the target value, can better learn the characteristics and laws of the data, and has good generalization ability. It does not combine the research object - quality and safety risk identification. 5. Conclusions As traditional risk identification methods cannot quickly and accurately identify the quality and safety of imported and exported toys, this paper puts forward a research framework of "data collection - risk classification - risk identification" for the quality and safety of imported and exported toys. Based on two unsupervised machine learning techniques, namely the LDA topic model and the WOA-BP neural network, a model for identifying the quality and safety risks of imported and exported toys is constructed. The proposed framework uses ROSTCM software to conduct TF-IDF frequency word processing on toy risk words. By constructing a network relationship graph, it is found that there is a correlation among high-frequency words. The LDA topic model is used to classify the high-frequency words by topics and divide them into toy safety risk factors and toy safety risk events. Meanwhile, the WOA-BP neural network and the BP neural network are constructed, and the data after topic division is brought into the neural network models. The results show that compared with the basic BP neural network model, the WOA-BP model shows higher accuracy and can better perform the task of identifying the quality and safety risks of imported and exported toys. The "data collection - risk classification - risk identification" framework constructed in this paper can also be extended to the identification of quality and safety risks of other imported and exported products. However, considering the dynamic development of imported and exported products and the complexity of international trade, this study only focuses on text content in data collection. In the future, factors such as multimodal data collection, feature marking of risk factors, and the complexity of the model should be considered to further optimize the model's algorithm and improve its applicability and accuracy. Declarations Author Contributions: Conceptualization, H.Q. and L.P.Z.; methodology, H.Q. and L.P.Z.; software, H.Q. and L.P.Z.; validation, H.Q. and L.P.Z.; formal analysis, H.Q. and L.P.Z.; investigation, H.Q. and L.P.Z.; resources, H.Q. and L.P.Z.; data curation, H.Q. and L.P.Z.; writing—original draft preparation, H.Q. and L.P.Z.; writing—review and editing, H.Q. and L.P.Z.; visualization, H.Q. and L.P.Z.; supervision, H.Q. and L.P.Z.; project administration, H.Q. and L.P.Z.; funding acquisition, H.Q. and L.P.Z. All authors have read and agreed to the published version of the manuscript. Data availability: The datasets used and/or analyses during the current study available from the corresponding author on reasonable request. Corresponding authors ： Pengzhao Li Funding: This work was supported by the Beijing Municipal Education Commission Research Plan General Project (grant number: KM202411232007) Acknowledgments: We are indebted to the anonymous reviewers and editor. Conflicts of Interest: The authors declare no conflict of interest.” References Aven, T. Risk assessment and risk management: Review of recent advances on their foundation. Eur. J. Oper. Res. 253 (1), 1–13 (2016). Wang, Y. L. & Zhan, J. L. An Exploration of U.S. Customs Risk Management. China Customs (03), 72–73. (2022). Wang, Y. L. Overview and Implications of French Customs Risk Management. China Customs (05), 86–87. (2024). Regmi, R. H. & Timalsina, A. K. Risk management in customs using deep neural network. In 2018 IEEE 3rd international conference on computing, communication and security (ICCCS) (pp. 133–137). (2018), October. Muralidharan, E., Hora, M. & Bapuji, H. Hazard severity and time to recall: Evidence from the toy industry. J. Bus. Res. 139 , 954–963 (2022). Cheng, G. Z., Li, T. Y. & Wang, G. P. Method for Identifying Driving Risks Based on the Spectrum of Dangerous Driving Behaviors on Icy and Snowy Roads. J. Transp. Syst. Eng. Inf. Technol. (04), 127–138. 10.16097/j.cnki.1009 (2024). – 6744.2024.04.013. Gupta, A. et al. Predicting stroke risk: An effective stroke prediction model based on neural networks. J. Neurorestoratology . 13 (1), 100156 (2025). Ovsyannikova, A. & Domashova, J. Identification of public procurement contracts with a high risk of non-performance based on neural networks. Procedia Comput. Sci. 169 , 795–799 (2020). Jiang, Y. B., Xu, N. W., Chen, T. X., Qin, A. C. & Huang, D. Z. Intelligent Classification and Recognition of Land Use and Simulation of Rainstorm and Flood Risks Based on BP Neural Network. J. Hebei Univ. (Natural Sci. Ed. ) (02), 208–215. (2024). Zhang, X. K., Jiang, Y., Wang, D. & Li, G. F. Research on the Constitutive Relationship of 25CrMo4 Steel Based on WOA-BP Neural Network. J. Plast. Eng. (08),182–187. (2023). Liu, C. F. & Zhang, G. Z. Research on Audit Risk Identification Based on SAE - BP Neural Network: Taking the Computer, Communication and Other Electronic Equipment Manufacturing Industry as an Example. Economic Probl. (06), 123–129. 10.16011/j.cnki.jjwt.2024.06.001 (2024). Zhao, X. T. & Hu, L. W. Research on the Inversion and Identification of Urban Traffic Congestion Point Sources Based on Typical Coupling Optimization Algorithms. J. Transp. Syst. Eng. Inf. Technol. (02), 74–83. 10.16097/j.cnki.1009-6744.2023.02.008 (2023). Ji, X. L., Zhang, H. H. & Han, S. Y. A merging approach for hole identification with the NMM and WOA-BP cooperative neural network in heat conduction problem. Eng. Anal. Boundary Elem. 169 , 106042 (2024). Li, Y. L., Wang, S. Q., Chen, Q. R. & Wang, X. G. Comparative Study of Several New Swarm Intelligence Optimization Algorithms. Comput. Eng. Appl. 22 , 1–12 (2020). Wang, G., Liu, M., Cao, D. & Tan, D. Identifying high-frequency–low-severity construction safety risks: an empirical study based on official supervision reports in Shanghai. Eng. Constr. Architectural Manage. 29 (2), 940–960 (2022). Liu, W. W., Wang, H. W., Ni, X. M., Hou, Z. G. & Peng, K. Research on the Identification of Civil Aviation Maintenance Risk Factors Based on the TG - LDA Model. Aeronaut. Comput. Technique (06), 45–49. (2023). Xu, S. Y., Chen, Q. Y. & Liu, P. F. Research on the Identification and Evolution of Important Risk Points in Equipment Manufacturing Enterprises Based on Sent - LDA. Mathematics in Practice and Theory, 1–18. Song, R. Z., Gao, X., Chen, H. Q. & Zeng, S. X. Research on the Ecological Risk Management of Railway Engineering Based on Machine Learning. Front. Eng. Manage. (01), 9–16. (2023). Wu, Y. L., Zhao, S. L., Li, C. J., Wei, N. D. & Wang, Z. Y. Text Classification Method Based on TF - IDF and Cosine Similarity. J. Chin. Inform. Process. (05), 138–145. (2017). Loh, S. L., Gan, C. K., Cheong, T. H., Salleh, S. & Sarmin, N. H. An Overview on Network Diagrams: Graph-Based Representation. J. Telecommunication Electron. Comput. Eng. (JTEC) . 8 (2), 83–86 (2016). Wang, J. F., Xu, Z., Feng, Y. C. & Zhang, H. Y. Identification and Evaluation of Technological Opportunities Based on Cluster Analysis and Multidimensional Space Patent Map. Industrial Eng. Manage. , 1–20 . Poggio, T., Mhaskar, H., Rosasco, L., Miranda, B. & Liao, Q. Why and when can deep-but not shallow-networks avoid the curse of dimensionality: a review. Int. J. Autom. Comput. 14 (5), 503–519 (2017). Zeng, X. E., Xue, J. J. & Huang, J. P. Construction of the Toy Safety Testing Standard System. Standard Sci. (09), 50–55. (2011). Ding, S., Su, C. & Yu, J. An optimizing BP neural network algorithm based on genetic algorithm. Artif. Intell. Rev. 36 , 153–162 (2011). Mirjalili, S. & Lewis, A. The whale optimization algorithm. Adv. Eng. Softw. 95 , 51–67 (2016). Ma, C., Zhou, D. Q. & Zhang, Y. Prediction Method of Water Resources Demand Based on BP Neural Network with Improved Whale Optimization Algorithm. Computer Science (S2), 486–490. (2020). Perera, R. et al. 1531 ‘Safety of toys: an unmet need in a developing country’ inquiry into safety of toys and parental knowledge on toy safety in Sri Lanka[J]. Archives Disease Child. 2021 , 106 (Suppl 1):A404–A405 . Hora, M., Bapuji, H. & ,Roth, V. A. .Safety hazard and time to recall: The role of recall strategy, product defect type, and supply chain player in the U.S. toy industry[J]. J. Oper. Manage. 2011 , 29 (7–8):766–777 . Seth, C. et al. The effects of innovation on product recall likelihood[J].Journal of Business Research,2024,173114452-. Kwong, C. W., Mak, S. L. & Li, C. H. Effectiveness of the Tactics for Small and Medium-sized Toy Factories in China in Dealing with European and US Toy Safety Requirements[C]//2021 IEEE International Conference on Industrial Engineering and Engineering Management (IEEM). IEEE, : 1179–1183. (2021). Mak, S. L. & Lau, H. K. An implementation of toy safety assessment model[C]//2014 IEEE Symposium on Product Compliance Engineering (ISPCE). IEEE, : 12–16. (2014). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 14 Jul, 2025 Reviews received at journal 13 Jul, 2025 Reviewers agreed at journal 12 Jul, 2025 Reviews received at journal 08 Jul, 2025 Reviewers agreed at journal 29 Jun, 2025 Reviewers agreed at journal 28 Jun, 2025 Reviewers agreed at journal 27 Jun, 2025 Reviewers agreed at journal 06 May, 2025 Reviewers invited by journal 06 May, 2025 Editor assigned by journal 06 May, 2025 Editor invited by journal 06 May, 2025 Submission checks completed at journal 02 May, 2025 First submitted to journal 10 Apr, 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-6416394","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":453532518,"identity":"b45d3e42-288e-4a8f-9723-c061c7a478d0","order_by":0,"name":"Qiong He","email":"","orcid":"","institution":"Beijing Information Science \u0026 Technology University","correspondingAuthor":false,"prefix":"","firstName":"Qiong","middleName":"","lastName":"He","suffix":""},{"id":453532519,"identity":"ce9b31ab-fcc2-4ebc-9612-91ffe3f7d053","order_by":1,"name":"Pengzhao Li","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4UlEQVRIie3QMUsDMRTA8ZRAuuTMmgNJv0Ig0FLoh3lBuEnBMUOHiOUyiHv7LW7smCNw03Pv2OIXaDc3dVfMuXXIb35/kvcIKYorxKYhRnArJWbv/RHcOp/ccLTxjI2pfXOnjzjkEyXB9LtNsp2/n9enDR3xMR4hVZ4aTeLcWc+ICC+Q2eUppmrP1GLim4Pd3xKJb92IV5Cb5TMZDhYZ0fIhk0jQqWql7YZJ+2hbOi7pd622HVJGxiUc4fvIYOotoxJw4NldZiGkM7hPJaS4XD7cWonw+nfyA//feFEURfGrL97ZUI4SQ9WdAAAAAElFTkSuQmCC","orcid":"","institution":"Beijing Information Science \u0026 Technology University","correspondingAuthor":true,"prefix":"","firstName":"Pengzhao","middleName":"","lastName":"Li","suffix":""}],"badges":[],"createdAt":"2025-04-10 04:53:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6416394/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6416394/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82562861,"identity":"2476391b-e3a9-4b76-8904-4f9d9502b166","added_by":"auto","created_at":"2025-05-13 01:50:49","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":136329,"visible":true,"origin":"","legend":"\u003cp\u003eFramework of the Quality and Safety Risk Identification Model\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6416394/v1/2dea9663a65ad051575b91d8.png"},{"id":82563288,"identity":"fbf94a55-9dee-424a-b7c7-3e9bc2a083ce","added_by":"auto","created_at":"2025-05-13 01:58:49","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2353237,"visible":true,"origin":"","legend":"\u003cp\u003eNetwork Relationship Diagram of High-Frequency Words\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6416394/v1/e497e1c1f78bf59ed0d0645b.png"},{"id":82562860,"identity":"1b554089-2baf-4c0c-be03-4f663f5c8dd3","added_by":"auto","created_at":"2025-05-13 01:50:49","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":35665,"visible":true,"origin":"","legend":"\u003cp\u003ePerplexity Corresponding to Different Numbers of Topics\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6416394/v1/3ab54c04c6570abc3c2f4cb4.png"},{"id":82562862,"identity":"62b47898-7c6d-4b3f-8a5f-b2893af84476","added_by":"auto","created_at":"2025-05-13 01:50:49","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":161480,"visible":true,"origin":"","legend":"\u003cp\u003eLDA Topic Clustering Results of Toy Safety Risks\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6416394/v1/9c0411b89b5a4bf9363ffba6.png"},{"id":82562256,"identity":"14897557-a92c-4c54-b6f8-634a0825f6b9","added_by":"auto","created_at":"2025-05-13 01:42:49","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":163225,"visible":true,"origin":"","legend":"\u003cp\u003eOptimization Flowchart of the WOA Algorithm\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6416394/v1/bc24789bd2b59b0584920f74.png"},{"id":82562863,"identity":"2f00025e-4968-46bb-8db8-5a09c1583427","added_by":"auto","created_at":"2025-05-13 01:50:49","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":239163,"visible":true,"origin":"","legend":"\u003cp\u003ePrediction Accuracy of Risk Events\u003c/p\u003e","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6416394/v1/7204b73f24df3393e1cf85ac.jpeg"},{"id":82562258,"identity":"6ffe2707-6ab8-464c-bb5a-cb3d194d7cea","added_by":"auto","created_at":"2025-05-13 01:42:49","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":177239,"visible":true,"origin":"","legend":"\u003cp\u003eRegression Analysis Graph\u003c/p\u003e","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6416394/v1/735d301c4edb77261e223e60.jpeg"},{"id":82563476,"identity":"1792e1a3-cb54-4f34-aadf-59d27aa58451","added_by":"auto","created_at":"2025-05-13 02:06:51","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3587019,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6416394/v1/29bef6ee-6217-4c8d-839f-2de0fcfc7598.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Research on Quality and Safety Risk Identification of Import and Export Toys Based on the WOA-BP Model","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eIn the field of international trade, risk management of the quality and safety of imported and exported products is a crucial link in ensuring the continuous, stable, and healthy development of trade activities[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Currently, different countries vary in their capabilities regarding the risk management of the quality and safety of imported and exported products. Among them, the US Customs is proficient in integrating resources in risk management and focuses on establishing an intelligence collection system, having developed automated identification systems such as ATS-G and ACE[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. France, on the other hand, conducts risk identification through three levels, namely the national, EU, and international levels, and uses big data algorithm technology to optimize the data pool[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. China's Customs combines import and export data and periodically selects cross-border e-commerce imported consumer goods that draw high consumer attention, have high quality and safety risks, and experience high growth rates in imports for quality inspections. Although different countries all have a complete set of approaches for the risk management of the quality and safety of imported and exported products, they are all constructed based on existing models and technological frameworks. When dealing with the increasingly complex and changeable international trade environment, there will be limitations, and it is difficult to accurately and efficiently identify all potential risks[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Stable and efficient risk identification technology is the foundation of risk management and can provide relevant standards for subsequent risk assessment and response. Toys, as special and sensitive products, are different from other imported and exported products, featuring compliance with diverse standards and regulations, as well as high-frequency contact and repeated use[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Therefore, this study selects imported and exported toys as the research object, constructs a set of quality and safety risk identification models for imports and exports, analyzes the risk factors of imported and exported toys, and identifies their quality and safety risks, with the aim of providing a novel method for the identification of quality and safety risks of imported and exported products.\u003c/p\u003e \u003cp\u003eRisk identification technology is a key link in the quality and safety risk management of import and export toys by customs. With the advent of the big data era, risk identification technology has also been continuously advancing. Currently, domestic and foreign scholars have applied neural network and other technologies in the field of risk identification. Guozhu Cheng constructed a dangerous behavior spectrum to classify the risk levels of dangerous driving behaviors and used the SVM statistical learning method to complete driving risk identification[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Aakanshi Gupta analyzed 8 machine learning algorithms such as support vector machines, random forests, and neural networks, collected relevant clinical data and used 8 machine algorithms to identify stroke risks. The results showed that the neural network model had high accuracy and normalization and stood out among these 8 machine learning algorithms[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Alexandra Ovsyannikova constructed a feedforward neural network model and trained it using batch normalization, specific weight initialization, and random gradient descent methods to solve the problem of contract default risk identification[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Yanbo Jiang constructed a land use classification model based on the BP neural network and combined it with a hydrological model and ArcGIS spatial analysis tools to identify flood risk[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Although neural network technology has achieved many results in the field of risk identification, its application also has some disadvantages[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. The currently popular BP neural network has problems such as slow learning speed, easy falling into local extrema, and uncertain values of network structure parameters when facing massive data. In response to this situation, Congfen Liu et al. used the SAE algorithm to optimize the BP neural network model to complete the research on audit risk identification, and the accuracy rate of the optimized model was greatly improved[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Xueting Zhao combined the principal component analysis method to optimize the BP neural network and improved the accuracy rate of the model for identifying vehicle operation risks in congested environments[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. X.L. Ji combined the NMM algorithm with the WOA-BP neural network to complete the identification of hole defects in heat conduction problems by detecting circular and elliptical hole defects[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. In addition, Yali Li et al. found that among several new intelligent optimization algorithms, the WOA algorithm has stronger adaptability, faster convergence speed, and simple parameter settings compared with other algorithms[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. When facing import and export commodity information with a large data scale, the WOA algorithm can identify potential quality and safety risk factors more quickly and comprehensively. In view of this, this study intends to use the WOA algorithm to optimize the BP neural network to solve problems such as the easy falling into local optimum and prediction instability of a single BP neural network, thereby improving the global optimization ability.\u003c/p\u003e \u003cp\u003eNeural networks have relatively strict requirements for input data. In the identification of import and export toy safety risks, toys come from a wide range of sources, and production standards vary in different countries and regions, resulting in diverse types of potential safety hazards in toys. The reasonable classification of toy information and an appropriate input set are of great significance for the accuracy of model results. Therefore, how to reasonably classify toy safety risk factors is the key to completing risk identification. In recent years, various fields have achieved rich results in risk factor classification. Wang et al. combined TF-IDF and Word2Vec and used the text clustering method to identify high-frequency and low-severity building safety report risks[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Weiwei Liu improved the LDA model by building a term dual optimization model, introduced the TF-IDF algorithm for text feature selection, optimized word weights using the Gaussian function, and obtained 26 risk topics for visualization by text clustering through the Sankey diagram[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Shangying Xu clustered risk topics of dual-source data sets based on the Sent-LDA model, identified 27 important risk points faced by equipment manufacturing enterprises, and performed discrete evolution analysis on the risk points[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Ruizhen Song et al. constructed a corpus, calculated the perplexity of text data, and identified railway engineering ecological risks based on the LDA \"topic-word\" distribution. Finally, 15 topic words with the largest correlation coefficients were obtained[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The LDA topic model can integrate and transform large-scale unquantifiable text data into structured data and extract valuable intelligence information. This study employs the LDA topic model to summarize the quality and safety text data of import and export toys into topics for classifying toy safety factors, facilitating the learning and training of the subsequent neural network model and enhancing the identification accuracy.\u003c/p\u003e \u003cp\u003eThe quality and safety risk identification of toys is the first step in risk management. Efficient and accurate risk identification can improve the risk management ability of import and export toy commodity quality and safety. In existing research, there is no objective and systematic identification of toy quality and safety risks. Based on this, this article takes the quality and safety of import and export toys as the research object and proposes a research framework of \"data collection-risk classification-risk identification\" based on two machine learning techniques: the topic model and the neural network.\u003c/p\u003e \u003cdiv id=\"Sec2\" class=\"Section2\"\u003e \u003ch2\u003e1.1 Construction of the Import and Export Quality and Safety Risk Identification Model\u003c/h2\u003e \u003cp\u003eAccording to the recall historical text data of import and export toy commodities, based on two machine learning techniques: topic analysis and neural network, a framework of the quality and safety risk identification model for import and export toys is constructed, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The model includes three parts: data collection, risk classification, and risk identification. In the data collection module, crawler technology is used to obtain recall information related to import and export toys from customs government websites. For the crawled text content, text cleaning, Chinese word segmentation, stop word removal, and text data initialization operations are carried out in sequence to complete text data preprocessing. The risk classification module includes two sub-modules: word frequency analysis and topic analysis. In the word frequency analysis sub-module, quality and safety keywords are first set, and then, based on ROSTCM software, keyword extraction is performed on the text information of import and export toys, TF-IDF word frequency statistics are carried out to form a correlation matrix, and the visualization processing of the word frequency matrix is completed. In the topic analysis sub-module, the LDA topic model is used to classify the screened keywords, screen out key topics and topic words, and divide toy safety risk factors and toy safety risk events based on this, providing data support for the subsequent model input and output environment. In the risk identification module, a quality and safety risk identification model for import and export toys is constructed based on the BP neural network model, and the WOA algorithm is used to optimize the parameters of the neural network to improve the accuracy of model identification.\u003c/p\u003e \u003cp\u003eThe model has three advantages. First, feature marking can simplify the workload. For a large amount of toy information, only the topic words after LDA topic analysis need to be feature marked to obtain the systematic import and export toy safety risk factors and toy safety risk events. Second, the word frequency analysis and topic analysis based on ROSTCM software and unsupervised machine learning models can avoid problems caused by the subjectivity of human clustering. Third, using the WOA algorithm to optimize the neural network model is expected to improve the overall identification accuracy of the model.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe organization of this paper is summarized as follows: Section 2 conducts data processing and LDA topic classification; Section \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e3\u003c/span\u003e constructs the neural network model; Section \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e4\u003c/span\u003e analyzes and compares the results of the model; and Section \u003cspan refid=\"Sec13\" class=\"InternalRef\"\u003e5\u003c/span\u003e summarizes this paper and discusses the directions for future research.\u003c/p\u003e \u003c/div\u003e "},{"header":"2. Classification of Quality and Safety Risk Factors of Import and Export Toys Based on LDA Model","content":"\u003ch2\u003e2.1 Data Collection\u003c/h2\u003e\n\u003cp\u003eThis article uses relevant government websites and customs data as data sources and uses Python\u0026apos;s data crawling technology to extract text data related to the recall of import and export toys. At the same time, this article extracts the EU RAPEX notification system, the US CPSC notification system, and the Canadian HC system from government websites and collects recall information about toys from 2010 to 2024. Among them, when screening the notification information of the US and Canada, keywords \u0026quot;toy\u0026quot; and \u0026quot;recall\u0026quot; are determined and screened in the title and abstract. Since the EU notification system reports toys and other commodities together, the request library is first used to obtain relevant reports in the web page, and then the content about toy recalls is extracted from the Beautifutlsup library to collect data from various countries. The text results are shown in Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e. Since the information contained in the title and keywords of toy recalls is limited, it may not be possible to obtain accurate toy risk types when used as data input into the topic model. Therefore, the toy defect content mentioned in the text is retained. 75% of the sample data is randomly selected from the database and divided into the training set, and the remaining 25% is used as the test set for subsequent model verification and analysis.\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eData Collection and Preprocessing Results\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDatabase\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRAPEX\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eCPCS\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eHC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eText Total\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTotal Search Texts\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3258\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e343\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e686\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4287\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eObtainable Texts\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3141\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e145\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e440\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3726\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTexts after Deleting Duplicates\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3130\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e138\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e374\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3642\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\n \u003ch2\u003e2.2 Word Frequency Analysis\u003c/h2\u003e\n \u003cp\u003eTF-IDF is a commonly used weighting technique for information retrieval and data mining, which can more accurately reflect the key degree of a word in a specific document and distinguish its importance weight[\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e]. ROSTCM is software mainly used for analyzing and processing text data. It uses natural language processing and text mining techniques to extract valuable information. In this study, based on ROSTCM software, TF-IDF word frequency analysis is performed on text data through functions such as dictionary setting, synonym setting, and part-of-speech screening. According to the analysis results, the top 30 high-frequency words ranked by TF-IDF values are listed, as shown in Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eHigh-Frequency Words in Import and Export Toy Recall Information\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eWord\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTF-IDF\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eWord\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTF-IDF\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eWord\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTF-IDF\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAsphyxiation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.056657\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDetachable\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.028335\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eStrangulation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.018003\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIngestion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.053896\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSoft toys\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.026622\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRubber toy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.014665\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlastic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.053784\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePollution\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.025059\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMagnetic flux\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.010894\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eReproductive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.048236\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFall off\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.022302\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSound pressure\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.010764\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSmall parts\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.046861\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlush toys\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.020829\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eElectric toys\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.010677\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlastic dolls\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.041271\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHearing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.019754\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRattle toys\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.010559\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePhthalic acid\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.039328\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBurns\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.020718\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFlammable\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.009698\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eToy clay\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.043076\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlastic toys\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.018515\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCuts\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.010238\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLaceration\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.033570\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFiber materials\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.018077\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMetal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.008908\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eComponents\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.031730\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eToy sets\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.017260\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIleus\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.008290\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003eThe network relationship diagram can show the complex relationship structure and understand the interrelationships between various entities[\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e]. Based on the word frequency matrix obtained after software processing, which reflects the correlation strength between words through the co-occurrence frequency of words, the matrix is visualized to obtain the network relationship diagram of high-frequency words in import and export toy recall information, as shown in Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003eCombined with the network diagram, the relationship analysis of some high-frequency words was conducted. \u0026quot;Asphyxiation\u0026quot; and \u0026quot;Ingestion\u0026quot; are in the core positions in the associated network. They are closely linked with \u0026quot;Small parts\u0026quot;, \u0026quot;Small components\u0026quot; and \u0026quot;Easy to fall off\u0026quot;, which means that toy parts are prone to falling off, putting children at risk of asphyxiation or accidentally ingesting them. Words related to \u0026quot;Plastic\u0026quot; and its derivatives frequently appear. Among them, \u0026quot;Plastic\u0026quot; is associated with \u0026quot; Contains phthalates\u0026quot;, and the latter is related to \u0026quot;Reproductive \u0026quot;, indicating that when plastic toys contain excessive amounts of this substance, it will endanger children\u0026apos;s reproductive health. \u0026quot;Toy clay\u0026quot; is closely associated with \u0026quot;Environmental pollution\u0026quot;, suggesting that toys may cause pollution to the environment during the use or disposal stages, resulting in their recall. Meanwhile, words like \u0026quot;Burns\u0026quot;, \u0026quot;Cuts\u0026quot; and \u0026quot;Strangulation\u0026quot; are presented around \u0026quot;Personal injury\u0026quot;, showing that the situations where toys endanger personal safety during the use process can also become the key factors for toy recalls. Through the above analysis, it can be found that there is a certain degree of correlation among different words. Although the frequency of occurrence of each word in the text and their mutual correlation have been clearly sorted out at the data level, the obtained output results are fragmented, and it is still unclear around which core themes the overall text unfolds. Therefore, it is necessary to classify the keys of the text and screen out the core themes and their corresponding theme words.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch2\u003e2.2 Word Frequency Analysis\u003c/h2\u003e\n\u003cp\u003eTF-IDF is a commonly used weighting technique for information retrieval and data mining, which can more accurately reflect the key degree of a word in a specific document and distinguish its importance weight[\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e]. ROSTCM is software mainly used for analyzing and processing text data. It uses natural language processing and text mining techniques to extract valuable information. In this study, based on ROSTCM software, TF-IDF word frequency analysis is performed on text data through functions such as dictionary setting, synonym setting, and part-of-speech screening. According to the analysis results, the top 30 high-frequency words ranked by TF-IDF values are listed, as shown in Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003eThe network relationship diagram can show the complex relationship structure and understand the interrelationships between various entities[\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e]. Based on the word frequency matrix obtained after software processing, which reflects the correlation strength between words through the co-occurrence frequency of words, the matrix is visualized to obtain the network relationship diagram of high-frequency words in import and export toy recall information, as shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003cbr\u003eCombined with the network diagram, the relationship analysis of some high-frequency words was conducted. \u0026quot;Asphyxiation\u0026quot; and \u0026quot;Ingestion\u0026quot; are in the core positions in the associated network. They are closely linked with \u0026quot;Small parts\u0026quot;, \u0026quot;Small components\u0026quot; and \u0026quot;Easy to fall off\u0026quot;, which means that toy parts are prone to falling off, putting children at risk of asphyxiation or accidentally ingesting them. Words related to \u0026quot;Plastic\u0026quot; and its derivatives frequently appear. Among them, \u0026quot;Plastic\u0026quot; is associated with \u0026quot; Contains phthalates\u0026quot;, and the latter is related to \u0026quot;Reproductive \u0026quot;, indicating that when plastic toys contain excessive amounts of this substance, it will endanger children\u0026apos;s reproductive health. \u0026quot;Toy clay\u0026quot; is closely associated with \u0026quot;Environmental pollution\u0026quot;, suggesting that toys may cause pollution to the environment during the use or disposal stages, resulting in their recall. Meanwhile, words like \u0026quot;Burns\u0026quot;, \u0026quot;Cuts\u0026quot; and \u0026quot;Strangulation\u0026quot; are presented around \u0026quot;Personal injury\u0026quot;, showing that the situations where toys endanger personal safety during the use process can also become the key factors for toy recalls. Through the above analysis, it can be found that there is a certain degree of correlation among different words. Although the frequency of occurrence of each word in the text and their mutual correlation have been clearly sorted out at the data level, the obtained output results are fragmented, and it is still unclear around which core themes the overall text unfolds. Therefore, it is necessary to classify the keys of the text and screen out the core themes and their corresponding theme words.\u003c/div\u003e\n\u003c/div\u003e\n\u003ch2\u003e2.3 Topic Analysis\u003c/h2\u003e\n\u003cp\u003eTo discover the hidden features, relationships, and patterns in the data and make a reasonable classification of risk factors, this paper extracts the information contained in the high-frequency words and network relationships screened by word frequency analysis through topic analysis. The topic model based on LDA is adopted to screen out the risk factors caused by toy quality and safety[\u003cspan class=\"CitationRef\"\u003e21\u003c/span\u003e]. LDA analysis is performed using the sklearn machine learning library in Python, and the optimal number of topics (K) is determined by perplexity. Perplexity is an effective method for evaluating and assisting in improving the parameters of a language probability model. By calculating the perplexity values corresponding to different numbers of topics and using the Python third-party toolkit Matplotlib to draw a graph, a line graph of the change in perplexity with the number of topics is obtained. When the perplexity is the lowest, the corresponding K value is the optimal. After analysis, when the K value is 4, the model\u0026apos;s perplexity is at a minimum value, as shown in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eAccording to the perplexity calculation, the number of topics in this study is determined to be 4. Subsequently, representative topic words are selected from the top 10 keywords with the highest frequency in the 4 groups of topic words, and the topic is named according to the topic words. The LDA topic clustering model is used for clustering analysis, and the clustering results are shown in Fig. \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e lists the top 30 most significant topic words related to the topic. According to the relevant standards of import and export toys, combined with the number of topics and keywords, they can be divided into toy types, toy materials, toy defects, and risk events. The specific topic words included in each topic are shown in Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eTopics and Corresponding Topic Words Divided by the LDA Topic Model\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTopic\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTopic Words\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eToy Types\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlush Toys, Electric Toys, Handmade Toys, Soft Toys, Plastic Toys, Rattle Toys, Decompression Toys, Kitchen Toys, Bath Toys, Magnetic Toys, Fishing Toys\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eToy Materials\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlastic, Wood, Plush, Metal, Paper, Rubber, Glass, Fiber Filling Materials\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eToy Defects\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEasily Fallen Off, Easily Detachable, Insufficient Stability, Easily Broken, Easily Fractured, Small Parts, Small Components, Sound Pressure Level, Flammable, Containing Excessive Phthalic Acid, High Magnetic Flux, Corrosion, Hygroscopicity, High Lead Concentration, High Cadmium Concentration, Sound Pressure Level, Dampness, Sharp Edges, Bacteria, Parasites, Warning Labels, Age Labels, Ingredient Labels, Instructions for Use, Safety Tips, Production Information\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInjury Events\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eReproductive System, Asphyxiation, Ingestion by Mistake, Personal Injury, Strangulation, Environmental Pollution, Hearing Impairment, Burns, Cuts, Intestinal Obstruction, Stabbing, Vision Impairment, Microbial Pollution, Chemical Hazards\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003ch2\u003e2.4 Classification of Safety Risk Factors\u003c/h2\u003e\n\u003cp\u003eThe LDA topic model, with its ability to deeply mine the semantic structure of text, extracts groups of words scattered throughout the text and related to different aspects of toys. Although it has initially outlined the contours of each topic, the relatively large number of topic words presents a certain degree of complexity and redundancy. In the subsequent neural network training process, this will lead to a high data dimension, triggering the \u0026quot;curse of dimensionality\u0026quot;[\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e]. Therefore, it is necessary to further summarize and integrate the topic words.\u003c/p\u003e\n\u003cp\u003eBased on the topic words and combined with the basic technical requirements to be complied with in the \u0026quot;Toy Safety Basic Specification,\u0026quot; the toy safety risk factors and toy safety risk events are classified. After topic analysis, a total of 14 toy safety risk events are screened out. Combined with the injuries caused by toy design defects, manufacturing processes, or materials listed in the specification, such as poisoning and other harmful substances injuries, asphyxiation, swallowing or inhaling foreign objects, electric shock, and other mechanical injuries including cuts, tears, abrasions, eye injuries, head injuries, and auditory injuries, these 14 safety risk events are classified and finally summarized into 5 categories: mechanical injuries, asphyxiation injuries, chemical injuries, hygienic injuries, and sensory injuries.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab4\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eSelection of Risk Feature Markers\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRisk Factor\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSpecific Description\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFeature Marker\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eToy Types\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlastic Toys, Electric Toys, Metal Toys, Doll Toys, Other Toys\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eA\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eToy Materials\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlastic Materials, Metal Materials, Textile Materials, Composite Materials, Other Materials\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eB\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eToy Defects\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePhysical Component Defects, Material Property Defects, Chemical Substance Defects, Biological Pollution Defects, Labeling Defects, Other Factors\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eC\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eAfter topic analysis, a total of 26 toy defects are screened out. Based on the mechanical physical safety testing standard system, the combustion safety testing standard system, the chemical element migration safety standard system, and the electromagnetic compatibility performance testing standard system in the toy safety testing standard system[\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e], the toy defects are divided according to the physical, chemical, biological, and labeling defects of the toy materials. Among them, factors with a word frequency of less than 5 times are classified as other factors. The factors that also affect toy risk identification include toy types and the materials of the toys themselves. Different toy types have differences in applicable ages and functional characteristics. After topic analysis, a total of 11 toy types are screened out. Based on the 3C toy certification classification in China, the toy types are divided into 5 categories, and factors with a word frequency of less than 5 times are classified as other factors. Different materials of toys will cause different safety risks. After topic analysis, a total of 8 toy materials are screened out and summarized into plastic types; metal types; rubber types; textile types; other materials. Feature markers are assigned to different risk factors. Toy types are marked as feature A; the material properties of the toys themselves are marked as feature B; and toy defects are marked as feature C, as shown in Table \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e. Through further division and integration of the topic words, redundant and interfering factors are eliminated to improve the accuracy and stability of the model in the prediction process. The feature-marked items are used as toy risk factors and applied to the input environment of the subsequent neural network, and the toy safety risk events are applied to the output environment of the subsequent neural network.\u003c/p\u003e"},{"header":"3. Construction of the Quality and Safety Risk Identification Model for Import and Export Toys","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Construction of the BP Neural Network Model\u003c/h2\u003e \u003cp\u003eThis study uses MATLAB software programming to implement the construction of the BP neural network model. The BP neural network model is usually composed of an input layer, a hidden layer, and an output layer. The features A, B, and C are used as the data source for the input layer of the BP neural network, and the number of input layer nodes is set to 3. The number of hidden layer nodes in the BP neural network has a great impact on the convergence speed and prediction accuracy of the network. If the number of nodes is too small, the network training error will be large, the iteration time will be long, and the identification accuracy of the predicted samples will be low. If the number of nodes is too large, the training time will be extended, and the training network will be overfitted[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Therefore, determining an appropriate number of hidden layer nodes is extremely critical. In this study, the number of hidden layer nodes is determined according to the formula (1):\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:M=\\sqrt{n+m}+c\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(1)\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\u003eIn formula (1), \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e is the number of input nodes; \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:m\\)\u003c/span\u003e\u003c/span\u003e is the number of output nodes; \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:c\\)\u003c/span\u003e\u003c/span\u003e is a constant from 1 to 10; \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:M\\)\u003c/span\u003e\u003c/span\u003e is the number of hidden layer nodes. According to the formula, the number of hidden layers in this study is determined to be 12. The output layer is set with 5 nodes, and each node corresponds to a category of toy risk events. Since there are 5 types of toy risk events, namely mechanical injuries, asphyxiation injuries, chemical injuries, hygienic injuries, and sensory injuries, during the training stage, the category labels are converted into one-hot encoded vector forms suitable for processing by the output layer of the neural network using the ind2vec function. After the prediction results are denormalized, the vector form output by the neural network is converted back into category labels using the vec2ind function to achieve correspondence and comparison with the original category labels.\u003c/p\u003e \u003cp\u003eIn addition, since each node in the input layer of this study contains different classifications and the parameter quantities of each node differ greatly, it is difficult to adjust the threshold values between layers, affecting the convergence rate and network accuracy. Therefore, the input parameters are normalized according to formula (2):\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabb\" border=\"1\"\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:y=\\frac{(x-min(X\\left)\\right)\\times\\:(ne{w}_{max}-ne{w}_{min})}{\\left(max\\right(X)-min(X\\left)\\right)}+ne{w}_{min}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(2)\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\u003eIn formula (2), \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:X\\)\u003c/span\u003e\u003c/span\u003e is the original data vector, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:min\\left(X\\right)\\)\u003c/span\u003e\u003c/span\u003e is the minimum value in \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:X\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:max\\left(X\\right)\\)\u003c/span\u003e\u003c/span\u003e is the maximum value in \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:X\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:ne{w}_{min}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:ne{w}_{max}\\)\u003c/span\u003e\u003c/span\u003e are the minimum and maximum values of the specified normalization interval (usually \u0026minus;\u0026thinsp;1 and 1). By uniformly normalizing the input parameters, the impact of data distribution differences on the generalization ability of the model can be reduced, so that the laws learned by the model on the training set can be better applied to the test set and practical applications.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Construction of the WOA-BP Network Model\u003c/h2\u003e \u003cp\u003eThe WOA algorithm simulates the group hunting behavior of whales and searches for the optimal solution through the update and search strategies of the whale's position[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. By combining the WOA algorithm with the BP neural network and using the global search ability of the WOA algorithm to optimize the weights and thresholds of the BP neural network, the performance of the BP neural network can be improved 26.\u003c/p\u003e \u003cp\u003eThe WOA algorithm calculates the fitness based on toy safety data. By applying the BP neural network under different parameter combinations to predict risk data and determining the fitness according to the degree of matching with the actual situation and comparing with the global optimal fitness after each calculation to update the global optimal position, which represents the current optimal parameter combination. In the WOA optimization algorithm, the random number p first divides the position update strategy of the search agent into individual selection-based or spiral upward. |A| further determines whether to approach the global optimal individual or a randomly selected individual under the selected type. When p\u0026thinsp;\u0026lt;\u0026thinsp;0.5 and |A| \u0026lt; 1, the search agent approaches the global optimal individual. If |A| \u0026ge; 1, the search agent approaches a randomly selected individual. When p\u0026thinsp;\u0026ge;\u0026thinsp;0.5, the search agent adopts a spiral update strategy. After each update, the fitness is recalculated, and the individual and global optimal positions are updated to promote the algorithm to evolve towards the optimal parameter combination. When the maximum iteration number is reached, the optimal parameter combination can be output, and then the training and prediction of the toy quality and safety risk identification model can be carried out. The optimization process of the WOA algorithm is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThis study defines a target function to measure the quality of each whale individual's position in the algorithm. When setting the target function, considering that the example is a multi-classification task, the way of minimizing the root mean square error is used for setting. The initial population size is 20, the maximum iteration number is 50, and the optimization parameter range is set between \u0026minus;\u0026thinsp;1 and 1.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Results Analysis","content":"\u003cp\u003eTo comprehensively evaluate the level of the WOA-BP model in identifying the quality and safety risks of import and export toys, the unoptimized BP model and the WOA-BP model are simulated and tested on the MATLAB 2022 platform. The ability of the WOA-BP model is judged through the accuracy analysis and regression analysis of the models.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Accuracy Performance Analysis\u003c/h2\u003e \u003cp\u003eThe prediction accuracy results of the BP model and the WOA-BP model for toy risk events are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e. It can be seen from the figure that in the test set, the prediction accuracy rate of the BP model is 90%, while the prediction accuracy rate of the WOA-BP model is 95.71%. The WOA-BP model has higher accuracy and smaller errors. Therefore, the prediction accuracy of the WOA-BP model is much better than that of the BP model, indicating that the WOA-BP model can better learn the characteristics and laws of the data and has good generalization ability.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Regression Analysis\u003c/h2\u003e \u003cp\u003eThe regression abilities of the BP model and the WOA-BP model are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e. The R value of the BP model's test set is 0.938, and the R value of the WOA-BP model's test set reaches 0.997. It can be seen from the figure that the regression effect of the WOA-BP test is better, and the fitting degree with the target value is higher. Overall, the WOA-BP test performs better than the BP test in this regression analysis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn summary, the WOA-BP model performs better than the BP model in predicting toy risk events. The WOA-BP model is more excellent in terms of accuracy and regression ability, has a higher fitting degree with the target value, can better learn the characteristics and laws of the data, and has good generalization ability. It does not combine the research object - quality and safety risk identification.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003eAs traditional risk identification methods cannot quickly and accurately identify the quality and safety of imported and exported toys, this paper puts forward a research framework of \"data collection - risk classification - risk identification\" for the quality and safety of imported and exported toys. Based on two unsupervised machine learning techniques, namely the LDA topic model and the WOA-BP neural network, a model for identifying the quality and safety risks of imported and exported toys is constructed.\u003c/p\u003e \u003cp\u003eThe proposed framework uses ROSTCM software to conduct TF-IDF frequency word processing on toy risk words. By constructing a network relationship graph, it is found that there is a correlation among high-frequency words. The LDA topic model is used to classify the high-frequency words by topics and divide them into toy safety risk factors and toy safety risk events. Meanwhile, the WOA-BP neural network and the BP neural network are constructed, and the data after topic division is brought into the neural network models. The results show that compared with the basic BP neural network model, the WOA-BP model shows higher accuracy and can better perform the task of identifying the quality and safety risks of imported and exported toys.\u003c/p\u003e \u003cp\u003eThe \"data collection - risk classification - risk identification\" framework constructed in this paper can also be extended to the identification of quality and safety risks of other imported and exported products. However, considering the dynamic development of imported and exported products and the complexity of international trade, this study only focuses on text content in data collection. In the future, factors such as multimodal data collection, feature marking of risk factors, and the complexity of the model should be considered to further optimize the model's algorithm and improve its applicability and accuracy.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAuthor Contributions:\u003c/strong\u003e Conceptualization, H.Q. and L.P.Z.; methodology, H.Q. and L.P.Z.; software, H.Q. and L.P.Z.; validation, H.Q. and L.P.Z.; formal analysis, H.Q. and L.P.Z.; investigation, H.Q. and L.P.Z.; resources, H.Q. and L.P.Z.; data curation, H.Q. and L.P.Z.; writing—original draft preparation, H.Q. and L.P.Z.; writing—review and editing, H.Q. and L.P.Z.; visualization, H.Q. and L.P.Z.; supervision, H.Q. and L.P.Z.; project administration, H.Q. and L.P.Z.; funding acquisition, H.Q. and L.P.Z. All authors have read and agreed to the published version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability:\u003c/strong\u003eThe datasets used and/or analyses during the current study available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCorresponding authors\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003ePengzhao Li \u0026lt;[email protected]\u0026gt;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e This work was supported by the Beijing Municipal Education Commission Research Plan General Project (grant number: KM202411232007)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments:\u003c/strong\u003e We are indebted to the anonymous reviewers and editor.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of Interest:\u0026nbsp;\u003c/strong\u003eThe authors declare no conflict of interest.”\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAven, T. Risk assessment and risk management: Review of recent advances on their foundation. \u003cem\u003eEur. J. Oper. Res.\u003c/em\u003e \u003cb\u003e253\u003c/b\u003e (1), 1\u0026ndash;13 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, Y. L. \u0026amp; Zhan, J. L. An Exploration of U.S. Customs Risk Management. \u003cem\u003eChina Customs\u003c/em\u003e (03), 72\u0026ndash;73. (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, Y. L. Overview and Implications of French Customs Risk Management. \u003cem\u003eChina Customs\u003c/em\u003e (05), 86\u0026ndash;87. (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRegmi, R. H. \u0026amp; Timalsina, A. K. Risk management in customs using deep neural network. In \u003cem\u003e2018 IEEE 3rd international conference on computing, communication and security (ICCCS)\u003c/em\u003e (pp. 133\u0026ndash;137). (2018), October.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMuralidharan, E., Hora, M. \u0026amp; Bapuji, H. Hazard severity and time to recall: Evidence from the toy industry. \u003cem\u003eJ. Bus. Res.\u003c/em\u003e \u003cb\u003e139\u003c/b\u003e, 954\u0026ndash;963 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCheng, G. Z., Li, T. Y. \u0026amp; Wang, G. P. Method for Identifying Driving Risks Based on the Spectrum of Dangerous Driving Behaviors on Icy and Snowy Roads. \u003cem\u003eJ. Transp. Syst. Eng. Inf. Technol.\u003c/em\u003e (04), 127\u0026ndash;138. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.16097/j.cnki.1009\u003c/span\u003e\u003cspan address=\"10.16097/j.cnki.1009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024). \u0026ndash;\u0026thinsp;6744.2024.04.013.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGupta, A. et al. Predicting stroke risk: An effective stroke prediction model based on neural networks. \u003cem\u003eJ. Neurorestoratology\u003c/em\u003e. \u003cb\u003e13\u003c/b\u003e (1), 100156 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOvsyannikova, A. \u0026amp; Domashova, J. Identification of public procurement contracts with a high risk of non-performance based on neural networks. \u003cem\u003eProcedia Comput. Sci.\u003c/em\u003e \u003cb\u003e169\u003c/b\u003e, 795\u0026ndash;799 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiang, Y. B., Xu, N. W., Chen, T. X., Qin, A. C. \u0026amp; Huang, D. Z. Intelligent Classification and Recognition of Land Use and Simulation of Rainstorm and Flood Risks Based on BP Neural Network. \u003cem\u003eJ. Hebei Univ. (Natural Sci. Ed.\u003c/em\u003e) (02), 208\u0026ndash;215. (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, X. K., Jiang, Y., Wang, D. \u0026amp; Li, G. F. Research on the Constitutive Relationship of 25CrMo4 Steel Based on WOA-BP Neural Network. \u003cem\u003eJ. Plast. Eng.\u003c/em\u003e (08),182\u0026ndash;187. (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, C. F. \u0026amp; Zhang, G. Z. Research on Audit Risk Identification Based on SAE - BP Neural Network: Taking the Computer, Communication and Other Electronic Equipment Manufacturing Industry as an Example. \u003cem\u003eEconomic Probl.\u003c/em\u003e (06), 123\u0026ndash;129. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.16011/j.cnki.jjwt.2024.06.001\u003c/span\u003e\u003cspan address=\"10.16011/j.cnki.jjwt.2024.06.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao, X. T. \u0026amp; Hu, L. W. Research on the Inversion and Identification of Urban Traffic Congestion Point Sources Based on Typical Coupling Optimization Algorithms. \u003cem\u003eJ. Transp. Syst. Eng. Inf. Technol.\u003c/em\u003e (02), 74\u0026ndash;83. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.16097/j.cnki.1009-6744.2023.02.008\u003c/span\u003e\u003cspan address=\"10.16097/j.cnki.1009-6744.2023.02.008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJi, X. L., Zhang, H. H. \u0026amp; Han, S. Y. A merging approach for hole identification with the NMM and WOA-BP cooperative neural network in heat conduction problem. \u003cem\u003eEng. Anal. Boundary Elem.\u003c/em\u003e \u003cb\u003e169\u003c/b\u003e, 106042 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, Y. L., Wang, S. Q., Chen, Q. R. \u0026amp; Wang, X. G. Comparative Study of Several New Swarm Intelligence Optimization Algorithms. \u003cem\u003eComput. Eng. Appl.\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e, 1\u0026ndash;12 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, G., Liu, M., Cao, D. \u0026amp; Tan, D. Identifying high-frequency\u0026ndash;low-severity construction safety risks: an empirical study based on official supervision reports in Shanghai. \u003cem\u003eEng. Constr. Architectural Manage.\u003c/em\u003e \u003cb\u003e29\u003c/b\u003e (2), 940\u0026ndash;960 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, W. W., Wang, H. W., Ni, X. M., Hou, Z. G. \u0026amp; Peng, K. Research on the Identification of Civil Aviation Maintenance Risk Factors Based on the TG - LDA Model. \u003cem\u003eAeronaut. Comput. Technique\u003c/em\u003e (06), 45\u0026ndash;49. (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu, S. Y., Chen, Q. Y. \u0026amp; Liu, P. F. Research on the Identification and Evolution of Important Risk Points in Equipment Manufacturing Enterprises Based on Sent - LDA. Mathematics in Practice and Theory, 1\u0026ndash;18.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong, R. Z., Gao, X., Chen, H. Q. \u0026amp; Zeng, S. X. Research on the Ecological Risk Management of Railway Engineering Based on Machine Learning. \u003cem\u003eFront. Eng. Manage.\u003c/em\u003e (01), 9\u0026ndash;16. (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu, Y. L., Zhao, S. L., Li, C. J., Wei, N. D. \u0026amp; Wang, Z. Y. Text Classification Method Based on TF - IDF and Cosine Similarity. \u003cem\u003eJ. Chin. Inform. Process.\u003c/em\u003e (05), 138\u0026ndash;145. (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLoh, S. L., Gan, C. K., Cheong, T. H., Salleh, S. \u0026amp; Sarmin, N. H. An Overview on Network Diagrams: Graph-Based Representation. \u003cem\u003eJ. Telecommunication Electron. Comput. Eng. (JTEC)\u003c/em\u003e. \u003cb\u003e8\u003c/b\u003e (2), 83\u0026ndash;86 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, J. F., Xu, Z., Feng, Y. C. \u0026amp; Zhang, H. Y. Identification and Evaluation of Technological Opportunities Based on Cluster Analysis and Multidimensional Space Patent Map. \u003cem\u003eIndustrial Eng. Manage.\u003c/em\u003e, 1\u0026ndash;20 .\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePoggio, T., Mhaskar, H., Rosasco, L., Miranda, B. \u0026amp; Liao, Q. Why and when can deep-but not shallow-networks avoid the curse of dimensionality: a review. \u003cem\u003eInt. J. Autom. Comput.\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e (5), 503\u0026ndash;519 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZeng, X. E., Xue, J. J. \u0026amp; Huang, J. P. Construction of the Toy Safety Testing Standard System. \u003cem\u003eStandard Sci.\u003c/em\u003e (09), 50\u0026ndash;55. (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDing, S., Su, C. \u0026amp; Yu, J. An optimizing BP neural network algorithm based on genetic algorithm. \u003cem\u003eArtif. Intell. Rev.\u003c/em\u003e \u003cb\u003e36\u003c/b\u003e, 153\u0026ndash;162 (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMirjalili, S. \u0026amp; Lewis, A. The whale optimization algorithm. \u003cem\u003eAdv. Eng. Softw.\u003c/em\u003e \u003cb\u003e95\u003c/b\u003e, 51\u0026ndash;67 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMa, C., Zhou, D. Q. \u0026amp; Zhang, Y. Prediction Method of Water Resources Demand Based on BP Neural Network with Improved Whale Optimization Algorithm. Computer Science (S2), 486\u0026ndash;490. (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePerera, R. et al. 1531 \u0026lsquo;Safety of toys: an unmet need in a developing country\u0026rsquo; inquiry into safety of toys and parental knowledge on toy safety in Sri Lanka[J]. \u003cem\u003eArchives Disease Child. 2021\u003c/em\u003e, \u003cb\u003e106\u003c/b\u003e(Suppl 1):A404\u0026ndash;A405 .\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHora, M., Bapuji, H. \u0026amp; ,Roth, V. A. .Safety hazard and time to recall: The role of recall strategy, product defect type, and supply chain player in the U.S. toy industry[J]. \u003cem\u003eJ. Oper. Manage. 2011\u003c/em\u003e, \u003cb\u003e29\u003c/b\u003e(7\u0026ndash;8):766\u0026ndash;777 .\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSeth, C. et al. The effects of innovation on product recall likelihood[J].Journal of Business Research,2024,173114452-.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKwong, C. W., Mak, S. L. \u0026amp; Li, C. H. Effectiveness of the Tactics for Small and Medium-sized Toy Factories in China in Dealing with European and US Toy Safety Requirements[C]//2021 IEEE International Conference on Industrial Engineering and Engineering Management (IEEM). IEEE, : 1179\u0026ndash;1183. (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMak, S. L. \u0026amp; Lau, H. K. An implementation of toy safety assessment model[C]//2014 IEEE Symposium on Product Compliance Engineering (ISPCE). IEEE, : 12\u0026ndash;16. (2014).\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":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Import and Export Toys, WOA-BP Neural Network, Risk Identification, Topic Model","lastPublishedDoi":"10.21203/rs.3.rs-6416394/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6416394/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eWithin the scope of customs risk management, the weakness in the ability to identify quality and safety risks of import and export products is a key factor contributing to the persistently high product recall rate. To effectively improve this situation, this paper selects import and export toy products as research samples and constructs a research system framework of \"data collection - risk classification - risk identification\". This study establishes a quality and safety risk identification model for import and export toys based on two machine learning techniques, namely Latent Dirichlet Allocation (LDA) and neural networks. Firstly, Use Python to preprocess the information and employ the ROSTCM software to conduct word frequency analysis to obtain a network relationship diagram. Based on the Dirichlet distribution topic model and toy safety-related indicators, keywords are extracted to determine toy safety risk factors as well as toy safety risk events. A safety risk identification model for import and export toys is established through the BP neural network, and the Whale Optimization Algorithm (WOA) is used to optimize the model. The results of the simulation study show that in terms of the model's accuracy, the prediction accuracy rate of the WOA-BP model is 95.71%, which is 5.71 percentage points higher than that of the BP model. In terms of the model's regression performance, the R-value of the WOA-BP model's test set is 0.997. The WOA-BP model is superior to the BP model and its prediction results are more in line with the actual situation, enabling it to better accomplish the task of identifying quality and safety risks.\u003c/p\u003e","manuscriptTitle":"Research on Quality and Safety Risk Identification of Import and Export Toys Based on the WOA-BP Model","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-13 01:42:44","doi":"10.21203/rs.3.rs-6416394/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-07-14T08:29:51+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-13T14:24:38+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"176516237836661204741068549149459754799","date":"2025-07-12T07:46:03+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-08T11:45:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"147834371449975402182025761395772437207","date":"2025-06-29T14:19:08+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"223113394124759335601090708739310503905","date":"2025-06-29T00:36:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"161391329570237848064000426930210501639","date":"2025-06-27T08:35:03+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"222241196224447836201976019343638848595","date":"2025-05-06T14:33:12+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-05-06T10:01:01+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-05-06T09:59:38+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-05-06T08:07:04+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-05-02T13:15:50+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-04-10T04:42:05+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"3cf566ed-c79d-4bd7-80f6-25498e9fbf27","owner":[],"postedDate":"May 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":48230821,"name":"Earth and environmental sciences/Environmental social sciences/Sustainability"},{"id":48230822,"name":"Physical sciences/Mathematics and computing/Computer science"},{"id":48230823,"name":"Physical sciences/Mathematics and computing/Information technology"},{"id":48230824,"name":"Physical sciences/Mathematics and computing/Statistics"}],"tags":[],"updatedAt":"2025-08-07T03:38:27+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-13 01:42:44","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6416394","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6416394","identity":"rs-6416394","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}