Identifying and Characterizing Eating Disorder Discourse on Chinese Social Media: A Machine Learning Approach

Identifying and Characterizing Eating Disorder Discourse on Chinese Social Media: A Machine Learning Approach | 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 Research Article Identifying and Characterizing Eating Disorder Discourse on Chinese Social Media: A Machine Learning Approach Yuchen Zhang, Nanyu Luo, Xiaoya Zhang, Feng Ji, Jinbo He This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7852043/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 8 You are reading this latest preprint version Abstract Background Eating disorders (EDs) are severe psychiatric conditions with high mortality and substantial medical complications. In China, underdiagnosis and low treatment engagement hinder timely intervention. Social media platforms provide a naturalistic lens into ED-related experiences, yet research on Chinese-language data remains scarce. Advances in machine learning (ML) and deep learning (DL) offer new opportunities to identify and characterize such ED discourse, informing the development of scalable detection methods and culturally tailored prevention and intervention strategies in the Chinese context. Methods We collected ED-related posts from Weibo via keyword-based API searches and manually annotated them into three groups: irrelevant, promotional/educational content, and layperson posts. Five ML/DL methods, including Convolutional Neural Networks (CNNs), Random Forests, XGBoost, Support Vector Machines (SVMs), and Logistic Regression, were trained to identify ED-related posts in a two-stage framework: (1) filtering out irrelevant posts and (2) distinguishing promotional/educational posts from layperson posts. Classifier performance was evaluated on additional posts from the same users. Latent Dirichlet Allocation (LDA) was applied to the layperson subset to extract underlying ED-related themes. Results CNN consistently outperformed other models, achieving high F1-scores in both classification stages (0.87 and 0.98, respectively). Topic modelling revealed five themes: restrictive symptomatology and physical distress, binge eating and body-health concerns, relapse and coping narratives, emotional venting, and chronic ED patterns with identity impact. Conclusions This study demonstrates that CNN-based classification combined with topic modeling provides a scalable framework for detecting ED-related discourse on Chinese social media. Beyond methodological advances in non-English NLP, the findings highlight culturally specific symptom expressions and psychosocial concerns, offering novel insights for public health surveillance. These insights can inform the development of early detection tools and culturally sensitive interventions to address the unmet needs of individuals with EDs in China. machine learning deep learning eating disorder topics topic modelling social media Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Eating disorders (EDs), such as anorexia nervosa (AN), bulimia nervosa (BN), and binge eating disorder (BED), are complex and severe psychiatric conditions marked by high mortality [ 1 – 3 ], and extensive adverse physical and psychological consequences, such as malnutrition [ 4 ]; metabolic disturbances [ 5 ]; comorbidity with major depression and generalized anxiety disorders [ 6 ]. EDs typically manifest during adolescence and early adulthood and often persist throughout life, bringing significant distress to both patients and their families [ 7 ]. Although incidence rates of EDs in China have been rising, their prevalence remains underestimated, and treatment engagement is low [ 8 – 10 ]. One plausible explanation is that individuals with EDs tend to exhibit self-denial and concealing behaviors for self-protection, which creates substantial challenges in assessing and treating these conditions [ 11 ]. Given China’s large population base and substantial unmet need, identifying risk factors and patterns of disordered eating is critical for prevention [ 12 ]. Yet, research on ED-related discourse in the Chinese context remains limited. Thus, there is an urgent need for scalable methods to capture risk patterns beyond clinical samples Social media provides a unique window into hidden or unreported mental health experiences [ 13 ], including EDs [ 14 – 15 ]. Platforms (e.g., Meta and Twitter/X in Western countries, and Weibo in China) have provided open and (semi-)anonymous spaces for people to share their daily thoughts and feelings, seek information, and discuss topics through various media (e.g., text, pictures, videos; [ 15 – 16 ]. These digital traces have been shown to reveal novel behavioral patterns and can complement epidemiological and clinical approaches [ 13 ]. In China, Weibo functions much like Twitter/X, which allows users to rapidly exchange information and form interactive social networks. Moreover, Chinese natural language processing (NLP) presents unique challenges, including word segmentation and idiomatic expressions [ 17 – 18 ]. Such linguistic and cultural features necessitate context-specific approaches rather than extrapolating from English-language research. Thus, rigorous study of Chinese platforms like Weibo is essential to uncover more culturally specific signals of EDs at the population scale. Moreover, because people in ED communities often share their thoughts, behaviors, and daily lives online [ 19 – 20 ], the large volumes of user-generated content on social media provide a strong basis for large-scale analysis in digital health [ 21 ]. Although Chinese datasets have linguistic and cultural features that differ from those of English, powerful toolkits, such as Machine Learning (ML), can still extract key signals useful for detecting risk factors and behavioral patterns [ 16 , 22 – 23 ]. In particular, advanced deep-learning models can accurately identify users showing ED-related symptoms from social media posts [ 10 , 20 ]. These methods help narrow broad user pools to specific at-risk groups, enabling closer examination of distinct patterns. ED-related topics are crucial to understanding the complexities of individuals’ mental health conditions. A retrospective literature review examined key trends in ED topics, including treatment, symptoms, and risk factors, from 1990 to 2021, emphasizing the importance of ongoing investigation in this field [ 24 ]. Moreover, researchers are also beginning to utilize social media data and big data analysis to enhance these insights [ 10 , 19 ]. For example, Wang and colleagues focused on detecting and characterizing ED communities on Twitter, which provided insights into community structures and interactions among ED individuals [ 20 ]. Machine learning can also further accelerate topic discovery at scale by minimizing the need for manual labeling. For instance, algorithms such as Latent Dirichlet Allocation (LDA) automatically extract latent themes related to social support and ED-specific content across large datasets, eliminating the need for intensive manual categorization that was previously required [ 25 ]. Similarly, Zhou and colleagues used the Correlation Explanation (CorEx) models to identify and cluster topics related to ED symptoms (e.g., body image, the consequences of EDs; [ 10 ]). Together, these methods demonstrate the potential of social media data to enhance prevention and treatment efforts in the public health sector. However, most existing work focuses on English-language datasets and evaluates a limited set of models. This leaves a gap in the understanding of culture-specific risk factors associated with EDs. Our study extends Zhou et al.’s method on Twitter data by adapting their ML classifiers and topic modeling methods to Weibo, thereby addressing the specific cultural and linguistic differences of Chinese-language social media content. Although Zhou et al. demonstrated the feasibility of English datasets, it remains unclear whether they apply equally well to other unique content [ 10 ]. Compared with Twitter, Weibo differs in most topic distributions and dynamics of social attention, as the entertainment context dominates the topics, and hashtag use also follows Weibo-specific conventions [ 26 ]. In this environment, body-image norms and shared dieting or eating habits can spread rapidly, often via celebrities, influencers, and trending tags, which might shape public perceptions and potentially amplify the risk of EDs. Additionally, Chinese natural language processing tasks require addressing word segmentation and Part-of-Speech tagging issues, which introduce challenges that are less prominent in English [ 27 ]. Moreover, ED-related topics in East Asian contexts are distinct from those in Western settings. For example, studies have found that Chinese young women reported higher levels of baseline stigmatization compared with Australian students (Yan et al., 2018). Additionally, research has demonstrated that East Asians are more likely to communicate negative experiences through somatic expression than Westerners [ 28 – 29 ]. Therefore, these specifications in cultural-linguistic features and platform-specific norms underscore the significance of examining ED-related content patterns in Chinese social media. Thus, we utilize data from Weibo to develop machine-learning techniques for exploring ED-related content in a culturally specific Chinese context. Specifically, our research addresses two key questions. The first step is to systematically assess the effectiveness of five distinct statistical and computational methods (e.g., Convolutional Neural Networks [CNNs], Random Forests [RFs], Extreme Gradient Boost [XGBoost], Support Vector Machines [SVMs], and Logistic Regression [LR]) in identifying ED-related content. The second research question aims at exploring and evaluating ED-related topics extracted from Weibo posts via Latent Dirichlet Allocation (LDA) so that risk factors or behavioural patterns associated with ED symptoms can be further revealed. By combining supervised classification with unsupervised topic modeling, we aim to demonstrate a scalable framework for ED surveillance in a non-English context. This approach not only advances methodological innovation but also provides culturally grounded insights into disordered eating, with potential applications for early detection and intervention in China. 2. Method 2.1 Study Design Our study methodology follows a two-stage pipeline shown in Fig. 1 . In the first stage, we collected an initial dataset of 6,589 Weibo posts containing keywords related to EDs via the Weibo Application Programming Interface (API). These posts are manually annotated into three categories (i.e., ED-laypeople, ED-promotion and education, and ED-irrelevant), which yields 3,469 posts in the ED-laypeople subset. Using these 3,469 posts as our training dataset, we developed and compared several traditional ML-classifiers alongside a deep learning (DL) model. The power of ML is utilized to identify ED-related posts and analyze ED-related topics within them, resulting in a twofold development. All models had two main tasks: first, to filter out ED-irrelevant posts and then to sort the remaining posts into either ED-promotion and education or ED-laypeople. Second, to evaluate the effectiveness of the classifiers, we identified the user IDs associated with the 3,469 laypeople posts and randomly sampled 20,000 additional posts from these same users as our testing datasets. Afterwards, the best-performing classifier retrieved 8,665 new laypeople subsets. In the second stage, we applied LDA topic modelling to both the manually labelled ED-laypeople subset (N = 3,469) and the classifier-identified subset (N = 8,665). The outcome of the topic modelling was scrutinized to assess how well it assisted in generating ED-relevant topics for each post. This was achieved by comparing each topic’s theme word to the content of posts under that topic, aiming to validate the relevance and accuracy of the automated topic modelling process. 2.2 Weibo Posts Collection This study was conducted in January 2023. We used the Weibo streaming API to search and collect posts relevant to our research. Utilizing a carefully selected list of keywords related to EDs (e.g., “厌食症 Anorexia”, “神经性厌食症 Anorexia Nervosa”, “神经性贪食 Bulimia Nervosa”, “贪食症 Bulimia”, “暴食症 Binge-eating Disorder”, “暴饮暴食 Binge eating”, “饮食失调 Eating disorder”, “狂食症 Compulsive overeating”, “进食障碍 Eating Disorder”, “易饿病 Excessive Hunger”) in Chinese, we were able to retrieve a comprehensive pool of posts for manual screening on Weibo. 2.3 Identification of Target Posts 2.3.1 Manually Annotated ED-laypeople Posts and ED-irrelevant Posts Our framework for identifying target posts was based on the “gold-standard” dataset distribution reported by Zhou et al., which annotated posts into three categories: ED-irrelevant, ED-promotional and education , and ED-laypeopl e. Posts that were unrelated to EDs were categorized as ED-irrelevant; posts aimed at promoting products or educating the public about EDs were classified as ED-promotional and education posts; and ED-laypeople posts referred to ED-relevant posts made by individual users. In this study, we focused specifically on ED-laypeople posts [ 10 ]. Three reviewers and two supervisors formed the annotation team for the manual review tasks in the study. All team members were pre-trained on the annotation principles to ensure consistency. Initially, the three reviewers independently annotated the posts. Then, they cross-checked their results to ensure the reliability of the results. Next, any inconsistencies were discussed with the supervisors to reach a final consensus on all posts. 2.3.2 Machine Learning Classifier Development To explore latent patterns within the manually labeled corpus, we employed ML algorithms to design automatic text classifiers. Since these algorithms only process numerical data, we pre-processed the texts. This involved removing elements that were potentially devoid of significant information, such as Chinese stop words (e.g., “但” [dàn], translating to “but”); user mentions (e.g., @username); hyperlinks (e.g., https://abccc.com ); hashtags (e.g., #gym); and emojis. The refined text was then segmented into individual words using the ‘ jieba ’ Chinese text segmentation toolkit. Subsequently, a comprehensive lexicon was established, assigning a unique numerical identifier to each word. This conversion transformed each post into a numeric vector representation. Our study centered on the construction of a classifier utilizing four conventional machine learning models, including Logistic Regression (LR), Support Vector Machines (SVMs), Random Forests (RFs), and Gradient Boosting Trees (XGBoost), alongside Convolutional Neural Networks (CNNs), a DL model. To enhance performance and avoid overfitting, we meticulously tuned the hyperparameters of the traditional models by amalgamating grid search with cross-validation. The CNNs’ structure consisted of three layers: an embedding layer, a convolutional block, and an output layer. The embedding layer, pre-populated with terms from both corpora mentioned above, facilitates the initial transformation of text into vector form. The convolutional block sequentially applied 1D convolutions with 64 filters of kernel size 3, followed by the ReLU activation function, max pooling with a kernel size of 2 for dimensionality reduction, and a dropout with a rate of 0.2 to deter overfitting. The last linear output layer was responsible for projecting the flattened features from the CNN onto the spectrum of class categories, yielding logits corresponding to each class. Our classification framework was designed in two stages. The initial phase entailed differentiating ED-related posts (including ED-layperson discussions and ED-promotion and education posts) from those irrelevant to ED, resulting in a binary classification. The ensuing stage engaged classifiers to distinguish between posts of ED-laypeople and those categorized under ED-promotion and education. 2.4 Identifying Topics from Posts of Potential ED Users Topic modeling techniques were used to extract topics and explore the mental status and daily concerns of potential ED users. Specifically, we utilized Latent Dirichlet Allocation (LDA; [ 30 ]), a probabilistic model designed to identify latent thematic structures within sizable textual datasets. This model assumes that documents are composed of a blend of latent topics, where each topic is defined by its unique word distribution. In practice, LDA interprets each document as a mixture of topics, assigning each word to a topic based on the distribution of words across topics and vice versa. We applied the LDA model to two distinct sets of data: (1) the initial dataset's posts that were manually labeled as ED-related discussions by laypeople and (2) specific posts from potential ED users, which were categorized as ED-laypeople discussions by our automated classifier. Determining the appropriate number of topics in LDA is a pivotal step. We experimented with different topic quantities, ranging from 4 to 10. While statistical measures such as perplexity and relevance scores offer guidance on the number of topics, they do not always yield thematically rich topics. Therefore, we conducted a series of testimonies and meticulously examined the resultant topics, allowing us to identify the most informative and representative topic number for subsequent in-depth analysis. 2.5 Topic evaluation Our team was instructed by domain experts to manually identify each topic from the top 30 most salient keywords generated by topic modeling and assign a corresponding key theme to each topic. Next, we then manually reviewed the top 10 most relevant posts associated with these keywords and evaluated whether the post was coherent with the generated topic theme. The relevance scores were defined as the percentages of posts that were coherent with each topic theme. 3. Results 3.1 Post Collection and Manual Identification of ED-related Content Using the keyword search feature via the Weibo API, we collected 6,589 posts. Of these, 3,489 were classified as discussions by ED laypeople, 1,338 as ED promotional and educational materials, and 1,782 as unrelated to EDs. The results of Cohen’s Kappa range from .58 to .79, indicating a moderate to substantial agreement among reviewers. The post collections of users whose posts were labeled as ED laypeople are the source of the second dataset, which consists of 20,000 posts randomly selected from these collections. 3.2 Machine Learning Classifier Development Our analysis encompassed five machine learning algorithms: CNNs, RFs, XGBoost, SVMs, and LR, mentioned above. The manually labeled dataset was randomly divided into a training set, encompassing 69.98% of the posts (4,611 out of 6,589), and a validation set, containing the remaining 30.02% (1,978 out of 6,589). Stratified sampling ensured equivalent distribution across both sets relative to the original dataset. We designed a two-tiered classification system to separate ED-irrelevant posts and to differentiate between ED-promotional and ED-laypeople discussions. Preliminary results indicate the classifiers' performances as shown in the subsequent Table 1 . Table 1 Performance comparison of classifiers under different thresholds Classifier Precision Recall F1-score Precision Recall F1-score ED-irrelevant versus other 2 labels (Threshold 0.5) ED-irrelevant (535) ED-promotional and education + ED-laypeople (1443) CNN 0.68 0.69 0.68 0.89 0.88 0.88 RF 0.83 0.55 0.66 0.85 0.96 0.9 XGBoost 0.81 0.51 0.63 0.84 0.96 0.9 SVM 0.68 0.36 0.47 0.8 0.94 0.86 LR 0.57 0.46 0.51 0.82 0.87 0.84 ED-irrelevant versus other 2 labels (Threshold 0.9) ED-irrelevant (535) ED-promotional and education + ED-laypeople (1443) CNN 0.64 0.75 0.69 0.9 0.85 0.87 RF 0.35 0.96 0.51 0.96 0.34 0.5 XGBoost 0.29 0.98 0.45 0.96 0.12 0.21 SVM 0.37 0.83 0.51 0.89 0.48 0.62 LR 0.31 0.92 0.46 0.89 0.25 0.39 ED-promotional and education versus ED-laypeople ED-promotional and education (402) ED-laypeople (1041) CNN 0.99 0.91 0.95 0.97 1 0.98 RF 0.95 0.92 0.93 0.97 0.98 0.97 XGBoost 0.91 0.87 0.89 0.95 0.97 0.96 SVM 0.87 0.75 0.8 0.91 0.96 0.93 LR 0.88 0.67 0.76 0.88 0.96 0.92 In the initial stage, we examined two methodologies to distinguish ED-irrelevant posts from the other two groups. For any post input, all the models were regarded as statistical models that output the probability of the input belonging to ED-irrelevant posts and the probability of the opposite. If the traditional 0.5 threshold is set, events with higher probability are more likely to occur, so the post should belong to the higher probability group. By setting the conventional threshold at 0.5, our models exhibited a commendable average F1-score of 0.87 but a modest average recall of 0.51 for the ED-irrelevant category, indicating that although the classifier successfully found ED-relevant posts, the posts predicted as no ED-irrelevant posts would be mixed with a considerable number of ED-irrelevant posts. Therefore, we set the threshold at a high value of 0.9, meaning the classifier considers the post ED-irrelevant only if the output probability exceeds 0.9. The adjustment of the threshold diminished the F1-scores for the traditional models; however, the CNN sustained its high F1-score (0.87) and enhanced the recall for ED-irrelevant posts to 0.75. During the second phase, the CNNs surpassed their traditional counterparts with an F1-score of 0.98. Consequently, we selected CNNs for automated text classification, which identified 8,665 posts as ED laypeople discussions within the second dataset of 20,000 posts of potential ED users. 3.3 Topic Modeling and Evaluation Applying the LDA model to the aforementioned corpora, 3,489 manually tagged ED laypeople posts and 8,665 CNN-classified ED laypeople posts, enabled topic extraction. Following extensive experimentation and analysis, we determined that five topics are optimal for clustering ED-related symptoms into corresponding ED topics. The emblematic themes associated with each topic are presented subsequently. The five topic themes that we extracted based on our manual labeling training data were (1) restrictive ED symptomatology & physical distress, (2) binge eating behaviors & body-health concerns, (3) ED episodes, relapse, & coping narratives, (4) emotional venting during ED episodes, and (5) chronic ED patterns & identity impact. The relevance scores for these topics were 90%, 90%, 90%, 80%, & 100%, respectively. According to the intertopic distance map (Fig. 2 ), topic theme 1 and topic theme 2 were positioned far apart, with large and non-overlapping bubbles, indicating two highly prevalent but lexically distinct subtopics. In contrast, topics 3, 4, and 5 formed a tight cluster with partial overlaps. This reflected shared topics related to ED episodes but with differentiated focuses. Specifically, topic theme 3 focused on diary-style and relapse-related language; topic theme 4 centered on affect-laden expression; and topic theme 5 emphasized the habitual-related impacts of chronic ED patterns (Table 2). Additional figures also showed the top-30 most relevant terms for each topic (Fig. 3 , 4 , 5 , 6 , 7 ). Table 2: Five topic themes and their 10 most relevant contents Topic Theme 1: restrictive ED symptomatology & physical distress Post Content keywords Keywords translation Relevance 1 饿得想吐, 单纯想吐, 厌食症倾向 nausea due to hunger/ plain nausea, anorexia tendency* 1 2 头晕眼花, 喘大气, 不想吃饭 dizzy, short of breath, no appetite 1 3 减肥,一天不吃还没感觉, 厌食症倾向 diet, no hunger after a day without food, anorexia tendency 1 4 饿/涨得睡不着, 暴饮暴食， 想吐 insomnia due to hunger/bloating; binge eating, want to vomit 1 5 新冠后遗症, 厌食症, 恶心, 瘦八斤 sequela of covid-19, anorexia*, nausea, lost 4kg 1 6 感觉得了厌食症, 吃饭像完成任务 feel I have anorexia, eating like a task 1 7 恶心,吃不下, 新冠后遗症，厌食症 nausea, no appetite, sequela of covid-19, anorexia 1 8 巨能吃,暴饮暴食 huge appetite, binge eating 0 9 吃一两口就饱, 吐, 怀疑有厌食症 full after several bites, vomit, suspect of anorexia 1 10 没饥饿感, 是不是有厌食症 No hunger, suspect of anorexia 1 Relevance Scores 90% * Anorexia tendency and anorexia means low appetite in China, incidating restrictive ED symptomology. People may use sentences like "I feels like I have gotten the anorexia that I have no appetite these days. Topic Theme 2: binge eating behaviors & body-health concerns Post Content keywords Keywords translation Relevance 1 暴饮暴食然后减肥, 熬夜 binge eating then diet, stay up 1 2 暴饮暴食两周多, 体重高 binge eating over 2 weeks, high weight 1 3 天天暴饮暴食 binge eating daily 1 4 暴饮暴食 binge eating 1 5 暴饮暴食 binge eating 1 6 严重的抑郁症、厌食症 severe depression, anorexia 0 7 夜间狂食症 night-eating syndrome 1 8 暴饮暴食诱发寻麻疹 binge eating triggered hives 1 9 免疫力差，躯体化反应，进食障碍 poor immune system, somatization reaction, ED 1 10 暴食症,执着进食 binge eating disorder, eating obsessively 1 Relevance Scores 90% Topic Theme 3: ED episodes, relapse, & coping narratives Post Content keywords Keywords translation Relevance 1 ***神经性贪食 f*** bulimia nervosa 0 2 打卡, 暴食症 meal log, binge eating disorders 1 3 暴食症的康复日记,进食障碍 bulimia recovery diary, ED 1 4 贪食症复发,心情很受影响 relapse of binge eating, mood affecting 1 5 暴食症, 考试 binge eating disorders, tests 1 6 稳定暴食欲望 stabilize cravings 1 7 贪食症 bulimia nervosa 1 8 今天没暴食, 不然又要后悔 no binge today, no regret 1 9 最近暴食症很严重…狂塞, 塞满胃部, 放弃思考 severe binge eating recently, stuffing stomach until full, give up thinking 1 10 暴食症的康复日记,气色不好, 停止暴食, 爱自己 binge eating recovery dairy, pale, stop binge, love yourself 1 Relevance Scores 90% Topic Theme 4: emotional venting during ED episodes Post Content keywords Keywords translation Relevance 1 厌食症 anorexia nervosa 0 2 像是得了厌食症, 生无可恋 like anorexia nervosa, nothing to live for 1 3 厌食症 anorexia nervosa 0 4 食べ過ぎて気持ち悪いっ,狂食症 I feel sick from eating too much, eating syndrome 1 5 狂食症的反例 Counterexample of binge eating disorders 0 6 厌食症, 狂食症, 心碎 Anorexia nervosa, eating syndrome, heart breaking 1 7 无语, 狂食症 I can’t even, eating syndrome 1 8 狂食症, ~! eating syndrome with excited tones 1 9 狂食症, !! eating syndrome with excited tones 1 10 狂食症, 真坑爹 eating syndrome, so unfair 1 Relevance Scores 70% Topic Theme 5: chronic ED patterns & identity impact Post Content keywords Keywords translation Relevance 1 兴趣爱好, 乱消费, 暴饮暴食, 一整天不出门 hobbies, overspending, binge eating, staying in bed all day 1 2 熬夜喝酒, 暴饮暴食, 早点猝死 Staying up late drinking, binge eating, planning to drop dead early 1 3 暴饮暴食, 排便 binge eating, bowel movement 1 4 狂食症 eating syndrome 1 5 暴饮暴食株, 撑的要死 strain, binge eating, feel unbearably stuffed 1 6 兴趣爱好, 乱消费, 暴饮暴食, 一整天不出门 hobbies, overspending, binge eating, staying in bed all day 1 7 兴趣爱好, 乱消费, 暴饮暴食, 一整天不出门 hobbies, overspending, binge eating, staying in bed all day 1 8 鸭梨山大, 狂食症 overwhelming stress, eating syndrome 1 9 狂吃嗜睡 eating syndrome, oversleeping 1 10 不快乐, 暴饮暴食 feeling down, binge eating 1 Relevance Scores 100% In addition to the five topic themes described above, our analysis also revealed several cross-cutting patterns in the post contents. Many posts linked ED symptoms with comorbid conditions (e.g., sleeping difficulties, anxiety, endocrine dyscrasia) and with bodily sensations (e.g., lowered immune function, urticaria, somatization). Furthermore, given the specific years represented in our dataset, a notable subset of posts referenced the sequelae of COVID-19. 4. Discussion This study developed a two-tier text classification and topic modeling pipeline to identify and characterize ED-related social media posts from laypeople. In Stage 1, the CNN maintained a strong F1-score (0.87). In Stage 2, the CNN outperformed traditional models and was used to classify 20,000 additional posts, subsequently identifying 8,665 posts from ED laypeople. LDA on the combined corpora yielded five interpretable topic themes related to EDs. Our CNN accurately classified social-media posts, showing that deep models learn predictive local phrase patterns that simpler baselines miss. This finding is consistent with reports that CNNs and other deep neural models perform strongly in text-based mental health detection [ 10 , 31 ]. In parallel, our topic modeling yielded coherent, clinically interpretable themes from unselected posts, adding information that complements the supervised classifier. Prior studies also show that LDA and related models reveal latent concerns, track shifts in discourse, and describe support dynamics in online mental health communities [ 25 , 32 ]. The distance between bubbles of Topic themes 1 and 2 suggests that restriction-focused ED symptomology (e.g., “no appetite”, “anorexia tendency”) is linguistically apart from binge-eating-focused ED symptoms (e.g., “binge eating”, weight complaints). The clustering of Topic themes 3, 4, and 5 indicates that laypeople often narrate their ED episodes alongside emotion dysregulation and personal consequences (e.g., staying overnight, overwhelming stress), such that episodic descriptions of EDs and coping strategies (e.g., recovery diaries, “staying in bed all day”) appear together. These findings align with the literature, which suggests that emotional dysregulation (e.g., emotional nonacceptance, emotional distress, poor impulse control, deficits in emotion awareness) is a central feature across various EDs [ 33 – 34 ]. Meanwhile, the presence of diary-style self-monitoring and logs aligns with the CBT-E practice, which encourages the use of a real-time self-monitoring mechanism to modify eating behaviors. In this setting, users can anonymously use social media to track their eating behaviors and support their self-regulation, while often seeking community as well, without disclosing their formal diagnosis or treatment. These five topic themes suggest a need for interventions that not only address cognitive reframing but also provide psychoeducation related to body-focused management (i.e., understanding hunger cues and techniques to soothe nausea), connect individuals to chronic recovery resources (i.e., routine-based self-care and narrative-based techniques), and ensure sustained long-term support. Our findings are based on ED-related content in user-generated posts and are therefore not suitable for clinical settings. However, as social media offers a medium to reach broader public audiences, including individuals who may conceal or underreport maladaptive eating behaviors, these data could help characterize behavioral patterns in those lay populations and inform targeted psychoeducation courses, interventions, and prevention programs. Additionally, using a stringent 0.9 threshold for the ED-irrelevant class trades precision for recall to balance the results, which may potentially place borderline and/or ambiguous posts in the ED-relevant stream (i.e., ED-promotional and educational content, ED-laypeople). Moreover, although overall relevance scores are high, the topic coherence measure remains partly subjective, which prevents interpretability and leaves room for refinement [ 35 ]. We could incorporate human-in-the-loop to adjust for those borderline cases to improve reliability in our future work. Further gains may be achieved by using more advanced models. Recurrent models such as LSTM can improve classification accuracy [ 31 ], and topic models like BERTopic [ 36 ] can better capture latent patterns in text. At the same time, social-media posts, especially on platforms such as Weibo, often contain noise, including slang, idioms, and ambiguous internet expressions. More careful preprocessing and cleaning, along with the use of higher-quality text sources, when possible, are therefore crucial for obtaining reliable and meaningful results. 5. Conclusion Our study provides actionable insights that CNN can reliably classify large volumes of social media posts and combine with LDA to cluster coherent themes that point to intervention targets. Practically, this approach can help public-health teams prioritize designs targeted to psychoeducational content for lay communities who may not disclose clinical diagnoses. For example, this offers a novel direction for implementing interventions focused on physical sensation, such as hunger cues, in EDs, and initiating long-term recovery management (i.e., routine-based recovery supports, narrative-based self-monitoring tools). For future work, we recommend incorporating human-in-the-loop review for borderline cases, utilizing more advanced topic and text models, refining platform-specific preprocessing, and validating findings against clinical samples to enhance reliability and inform the design of intervention trials more effectively. Abbreviations AN Anorexia Nervosa BED Binge Eating Disorder BN Bulimia Nervosa CNN Convolutional Neural Network DL Deep Learning ED Eating Disorder LDA Latent Dirichlet Allocation LR Logistic Regression ML Machine Learning NLP Natural Language Processing RF Random Forest SVM Support Vector Machine XGBoost Extreme Gradient Boosting Table 2 Five topic themes and their 10 most relevant contents Declarations Ethical approval and consent to participate Ethical approval The project was approved by the Institutional Review Board (Applied Psychology) of The Chinese University of Hong Kong, Shenzhen (Approval No. EF20241018001). Informed consent Not Applicable. Consent for publication Not applicable. Availability of data and materials The data are available from the corresponding author upon request. Competing Interest The authors(s) declare that they have no competing interests. Funding Jinbo He was supported by the National Natural Science Foundation of China (Grant Number 72204208). Feng Ji was supported by the Connaught Fund (Grant Number 520245) and Social Sciences and Humanities Research Council (SSHRC) of Canada (Grant Number 215119, CRC-2024-00169). Authors’ Contributions Yuchen Zhang: Investigation, Project administration, Writing – original draft, Writing – review & editing; Nanyu Luo: Formal analysis, Writing – original draft, Writing – review & editing; Xiaoya Zhang: Writing – review & editing, Methodology; Feng Ji: Conceptualization, Supervision, Funding acquisition, Project administration, and Writing – review & editing; Jinbo He: Conceptualization, Supervision, Funding acquisition, Project administration, and Writing – review & editing. Declaration of generative AI in scientific writing During the preparation of this work, the authors utilized GPT-5 to enhance the language of the manuscript. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article. Acknowledgements Not Applicable. References Klump KL, et al. Academy for eating disorders position paper: eating disorders are serious mental illnesses. Int J Eat Disord. 2009;42(2). http://doi.org/10.1002/eat.20589 . Krug I et al. A meta-analysis of mortality rates in eating disorders: an update of the literature from 2010 to 2024. Clinical psychology review, 2025: p. 102547. https://doi.org/10.1016/j.cpr.2025.102547 Smink FR, Van Hoeken D, Hoek HW. Epidemiology of eating disorders: incidence, prevalence and mortality rates. Curr psychiatry Rep. 2012;14(4):406–14. https://doi.org/10.1007/s11920-012-0282-y . Carlson JL, Lemly DC. Medical Considerations and Consequences of Eating Disorders. Focus. 2024;22(3):301–6. https://doi.org/10.1176/appi.focus.20230042 . Mitchell JE. Medical comorbidity and medical complications associated with binge-eating disorder. Int J Eat Disord. 2016;49(3):319–23. https://doi.org/10.1002/eat.22452 . Keski-Rahkonen A, Mustelin L. Epidemiology of eating disorders in Europe: prevalence, incidence, comorbidity, course, consequences, and risk factors. Curr Opin Psychiatry. 2016;29(6):340–5. http://doi.org/10.1097/YCO.0000000000000278 . Chavez M, Insel TR. Eating disorders: National Institute of Mental Health's perspective. Am Psychol. 2007;62(3):159. https://doi.org/10.1037/0003-066X.62.3.159 . Li Z, et al. Burden of eating disorders in China, 1990–2019: an updated systematic analysis of the global burden of disease study 2019. Front Psychiatry. 2021;12:632418. https://doi.org/10.3389/fpsyt.2021.632418 . Qian J et al. An update on the prevalence of eating disorders in the general population: a systematic review and meta-analysis. Eating and Weight Disorders-Studies on Anorexia, Bulimia and Obesity, 2022. 27(2): pp. 415–28. https://doi.org/10.1007/s40519-021-01162-z Zhou S, et al. Exploring eating disorder topics on Twitter: machine learning approach. JMIR Med Inf. 2020;8(10):e18273. http://doi.org/10.2196/18273 . Howard LM, et al. The role of denial in eating disorder development, assessment, and treatment , in Eating Disorders . Springer; 2023. pp. 367–83. Wu J, et al. Secular trends in the incidence of eating disorders in China from 1990 to 2017: a joinpoint and age–period–cohort analysis. Psychol Med. 2022;52(5):946–56. https://doi.org/10.1017/S0033291720002706 . De Choudhury M et al. Predicting depression via social media . in Proceedings of the international AAAI conference on web and social media . 2013. https://doi.org/10.1609/icwsm.v7i1.14432 Chancellor S et al. Quantifying and predicting mental illness severity in online pro-eating disorder communities . in Proceedings of the 19th ACM conference on computer-supported cooperative work & social computing . 2016. https://doi.org/10.1145/2818048.2819973 Derenne J, Beresin E. Body image, media, and eating disorders—a 10-year update. Acad Psychiatry. 2018;42(1):129–34. https://doi.org/10.1007/s40596-017-0832-z . Wang X, et al. Depression risk prediction for chinese microblogs via deep-learning methods: Content analysis. JMIR Med Inf. 2020;8(7):e17958. http://doi.org/10.2196/17958 . Sproat R, Emerson T. The first international Chinese word segmentation bakeoff . in Proceedings of the second SIGHAN workshop on Chinese language processing . 2003. Zheng C, Huang M, Sun A. ChID: A large-scale Chinese IDiom dataset for cloze test. arXiv preprint arXiv:1906.01265, 2019. Park E, Kim W-H, Kim S-B. What topics do members of the eating disorder online community discuss and empathize with? an application of big data analytics . in Healthcare . 2022. MDPI. https://doi.org/10.3390/healthcare10050928 Wang T et al. Detecting and characterizing eating-disorder communities on social media . in Proceedings of the Tenth ACM International conference on web search and data mining . 2017. https://doi.org/10.1145/3018661.3018706 Mathews SC, et al. Digital health: a path to validation. NPJ Digit Med. 2019;2(1):38. https://doi.org/10.1038/s41746-019-0111-3 . Guan L, et al. Identifying Chinese microblog users with high suicide probability using internet-based profile and linguistic features: classification model. JMIR mental health. 2015;2(2):e4227. Huang X et al. Detecting suicidal ideation in Chinese microblogs with psychological lexicons . in. 2014 IEEE 11th Intl Conf on Ubiquitous Intelligence and Computing and 2014 IEEE 11th Intl Conf on Autonomic and Trusted Computing and 2014 IEEE 14th Intl Conf on Scalable Computing and Communications and Its Associated Workshops . 2014. IEEE. http://doi.org/10.1109/UIC-ATC-ScalCom.2014.48 Park E, Kim W-H. A retrospective literature review of eating disorder research (1990–2021): application of bibliometrics and topical trends. Int J Environ Res Public Health. 2022;19(13):7710. https://doi.org/10.3390/ijerph19137710 . Moessner M, et al. Analyzing big data in social media: Text and network analyses of an eating disorder forum. Int J Eat Disord. 2018;51(7):656–67. https://doi.org/10.1002/eat.22878 . Zhang Q, Gonçalves B. Topical differences between chinese language Twitter and sina weibo . in Proceedings of the 25th International Conference Companion on World Wide Web . 2016. https://doi.org/10.1145/2872518.2890562 Xu Q, Wang Z. A Data-Driven Model for Automated Chinese Word Segmentation and POS Tagging. Comput Intell Neurosci. 2022;2022(1):7622392. https://doi.org/10.1155/2022/7622392 . Choi E, Chentsova-Dutton Y, Parrott WG. The effectiveness of somatization in communicating distress in Korean and American cultural contexts. Front Psychol. 2016;7:383. https://doi.org/10.3389/fpsyg.2016.00383 . Ryder AG, et al. The cultural shaping of depression: somatic symptoms in China, psychological symptoms in North America? J Abnorm Psychol. 2008;117(2):300. http://doi.org/10.1037/0021-843X.117.2.300 . Blei DM, Ng AY, Jordan MI. Latent dirichlet allocation. J Mach Learn Res. 2003;3(Jan):993–1022. Kour H, Gupta MK. An hybrid deep learning approach for depression prediction from user tweets using feature-rich CNN and bi-directional LSTM. Multimedia Tools Appl. 2022;81(17):23649–85. https://doi.org/10.1007/s11042-022-12648-y . Low DM, et al. Natural language processing reveals vulnerable mental health support groups and heightened health anxiety on reddit during covid-19: Observational study. J Med Internet Res. 2020;22(10):e22635. http://doi.org/10.2196/22635 . Gratz KL, Roemer L. Multidimensional assessment of emotion regulation and dysregulation: Development, factor structure, and initial validation of the difficulties in emotion regulation scale. J Psychopathol Behav Assess. 2004;26(1):41–54. https://doi.org/10.1023/B:JOBA.0000007455.08539.94 . Lavender JM, et al. Dimensions of emotion dysregulation in bulimia nervosa. Eur Eat Disorders Rev. 2014;22(3):212–6. http://doi.org/10.1016/j.cpr.2015.05.010 . Röder M, Both A, Hinneburg A. Exploring the space of topic coherence measures . in Proceedings of the eighth ACM international conference on Web search and data mining . 2015. https://doi.org/10.1145/2684822.2685324 Grootendorst M. BERTopic: Neural topic modeling with a class-based TF-IDF procedure. arXiv preprint arXiv:2203.05794, 2022. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 05 May, 2026 Reviews received at journal 26 Mar, 2026 Reviewers agreed at journal 06 Mar, 2026 Reviewers agreed at journal 23 Nov, 2025 Reviewers invited by journal 02 Nov, 2025 Editor assigned by journal 30 Oct, 2025 Submission checks completed at journal 30 Oct, 2025 First submitted to journal 13 Oct, 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-7852043","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":538782408,"identity":"c31dbab6-187f-4d61-b586-e9760affb562","order_by":0,"name":"Yuchen Zhang","email":"","orcid":"","institution":"University of Toronto","correspondingAuthor":false,"prefix":"","firstName":"Yuchen","middleName":"","lastName":"Zhang","suffix":""},{"id":538782409,"identity":"a97ce5ea-28c1-480d-9bf0-3270bf2ff503","order_by":1,"name":"Nanyu Luo","email":"","orcid":"","institution":"University of Toronto","correspondingAuthor":false,"prefix":"","firstName":"Nanyu","middleName":"","lastName":"Luo","suffix":""},{"id":538782410,"identity":"fed68cf0-a75d-4aca-833e-3413deaa18fb","order_by":2,"name":"Xiaoya Zhang","email":"","orcid":"","institution":"University of Florida","correspondingAuthor":false,"prefix":"","firstName":"Xiaoya","middleName":"","lastName":"Zhang","suffix":""},{"id":538782411,"identity":"fa16d4fc-a4ed-43d3-b232-1012492e111e","order_by":3,"name":"Feng Ji","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAArklEQVRIiWNgGAWjYBAC9gaGhAMfSNLCc4Ah4eAMBgMQm7GBWC0MzDykaWFgeHjYtu2PnHl77/EHHxjs5InRknA4t83AWObMucTGGQzJhgStsodqSZwhkWPYzMNwgLDrwLZYImmxJ04LI5KWRMJamIGB3HPO2FiC54zhzBkGycmEtbD3JH/4USYnJ8HeY/DhQ4WdLUEtwEhJQOIZEFQPAuwHiFI2CkbBKBgFIxgAAKE6N3jv0/hxAAAAAElFTkSuQmCC","orcid":"","institution":"University of Toronto","correspondingAuthor":true,"prefix":"","firstName":"Feng","middleName":"","lastName":"Ji","suffix":""},{"id":538782412,"identity":"c1337df1-457b-44f4-98e6-4593128b4a73","order_by":4,"name":"Jinbo He","email":"","orcid":"","institution":"Xi’an Jiaotong-Liverpool University","correspondingAuthor":false,"prefix":"","firstName":"Jinbo","middleName":"","lastName":"He","suffix":""}],"badges":[],"createdAt":"2025-10-13 19:53:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7852043/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7852043/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":95802192,"identity":"68ca4900-6822-4607-843b-a419273eefcf","added_by":"auto","created_at":"2025-11-13 08:27:06","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":933775,"visible":true,"origin":"","legend":"","description":"","filename":"topicmodeling251027.docx","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/4bf9ed5782c03dbff2ec0ffe.docx"},{"id":95748415,"identity":"574f9a1f-82f1-48a3-b010-0d5354f3061e","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7741,"visible":true,"origin":"","legend":"","description":"","filename":"6ce410a6f289408d9cdfc97804bf5092.json","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/4adfccdb39f381344482e2c3.json"},{"id":95748416,"identity":"a8c680f4-c26d-40bb-a1da-8895aa4432be","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":123425,"visible":true,"origin":"","legend":"","description":"","filename":"6ce410a6f289408d9cdfc97804bf50921enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/bd11b71ee7a461921b26fce8.xml"},{"id":95748421,"identity":"1d5c105f-f3c9-48f6-85bc-e9db10d8a011","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":36909,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/ea5e4556b8b40cff812e96c0.png"},{"id":95748428,"identity":"f4fa46ad-2537-4745-90e6-131c76b65e9b","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":79422,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/6fdc3fb9f5a70da05717816b.png"},{"id":95748423,"identity":"ebd0f9c7-dc04-470b-a0c4-12a35f4fcc6a","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":88648,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/f84d3a1d26498365aab2cc86.png"},{"id":95748426,"identity":"fc813fee-1412-4d71-9251-84f9b12ed937","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":56794,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/c05b255d8a11d0963080ce53.png"},{"id":95801707,"identity":"4f139222-dbce-4ede-b4b2-a1079b2d4c67","added_by":"auto","created_at":"2025-11-13 08:26:00","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":58960,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/d44d8d5ce5d5a9b8c0ea772c.png"},{"id":95801367,"identity":"5a2b4fba-3ee4-48df-b52e-f2957cc444c6","added_by":"auto","created_at":"2025-11-13 08:25:11","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":94842,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/6884aa903650d5c60b5bddb8.png"},{"id":95800042,"identity":"3026c972-2cdb-4347-9c62-cd532d6efe0c","added_by":"auto","created_at":"2025-11-13 08:21:26","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":45845,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/d7d1364f59936210d404f059.png"},{"id":95748429,"identity":"0084306e-93bf-444e-981f-ef4986901f33","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"xml","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":124255,"visible":true,"origin":"","legend":"","description":"","filename":"6ce410a6f289408d9cdfc97804bf50921structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/1adb1554f5973ce7c8284184.xml"},{"id":95748431,"identity":"92c9eaf3-25a7-466e-9021-9fe659513d21","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"html","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":131876,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/51496f516dd4f20a3b63934f.html"},{"id":95748414,"identity":"e397eca2-04ec-4c59-9086-e4d6f8978a6b","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":67674,"visible":true,"origin":"","legend":"\u003cp\u003eA two-stage pipeline of our study methodology\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/44e58e8d51cc0da2d741bbdb.png"},{"id":95748419,"identity":"637847d9-33ee-475b-880d-708faf300475","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":272618,"visible":true,"origin":"","legend":"\u003cp\u003eIntertopic Distance Map for Labeling Dataset Preview\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/3d056511c390b62125abeb76.jpeg"},{"id":95748425,"identity":"0e3accaf-9534-433e-bb1a-cd2fff91315d","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":283520,"visible":true,"origin":"","legend":"\u003cp\u003eIntertopic Distance Map for Labeling Dataset (Top-30 most relevant terms for Topic 1)\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/7f4c5ce2fe2fe0ebbf3ef35b.jpeg"},{"id":95802230,"identity":"f09a71b9-f5ef-4489-8427-deaa7f340b8a","added_by":"auto","created_at":"2025-11-13 08:27:13","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":296427,"visible":true,"origin":"","legend":"\u003cp\u003eIntertopic Distance Map for Labeling Dataset (Top-30 most relevant terms for Topic 2)\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/0d16261ae04fefff1fd3d067.jpeg"},{"id":95801709,"identity":"fd57efb0-b390-44e6-b3cb-f855eb6be9f4","added_by":"auto","created_at":"2025-11-13 08:26:00","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":320657,"visible":true,"origin":"","legend":"\u003cp\u003eIntertopic Distance Map for Labeling Dataset (Top-30 most relevant terms for Topic 3)\u003c/p\u003e","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/8ee0e70e6efed9c104ed9382.jpeg"},{"id":95801134,"identity":"cc6c8922-2512-4688-a4a9-1a1cbfb30c5f","added_by":"auto","created_at":"2025-11-13 08:24:32","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":316896,"visible":true,"origin":"","legend":"\u003cp\u003eIntertopic Distance Map for Labeling Dataset (Top-30 most relevant terms for Topic 4)\u003c/p\u003e","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/7028d551338359407d8db327.jpeg"},{"id":95748417,"identity":"51750942-b4f7-4db7-ba43-5ba9651efe98","added_by":"auto","created_at":"2025-11-12 15:14:35","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":119031,"visible":true,"origin":"","legend":"\u003cp\u003eIntertopic Distance Map for Labeling Dataset (Top-30 most relevant terms for Topic 5)\u003c/p\u003e","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/1bef61a6ddbccf9bb23af847.jpeg"},{"id":95805250,"identity":"ae6454c0-586f-44c5-8cf4-019879b7671f","added_by":"auto","created_at":"2025-11-13 08:40:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2678818,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7852043/v1/4ff27c16-3ea2-4e3e-a59f-b3e059ce202b.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Identifying and Characterizing Eating Disorder Discourse on Chinese Social Media: A Machine Learning Approach","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eEating disorders (EDs), such as anorexia nervosa (AN), bulimia nervosa (BN), and binge eating disorder (BED), are complex and severe psychiatric conditions marked by high mortality [\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], and extensive adverse physical and psychological consequences, such as malnutrition [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]; metabolic disturbances [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]; comorbidity with major depression and generalized anxiety disorders [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. EDs typically manifest during adolescence and early adulthood and often persist throughout life, bringing significant distress to both patients and their families [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eAlthough incidence rates of EDs in China have been rising, their prevalence remains underestimated, and treatment engagement is low [\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. One plausible explanation is that individuals with EDs tend to exhibit self-denial and concealing behaviors for self-protection, which creates substantial challenges in assessing and treating these conditions [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Given China\u0026rsquo;s large population base and substantial unmet need, identifying risk factors and patterns of disordered eating is critical for prevention [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Yet, research on ED-related discourse in the Chinese context remains limited. Thus, there is an urgent need for scalable methods to capture risk patterns beyond clinical samples\u003c/p\u003e\u003cp\u003eSocial media provides a unique window into hidden or unreported mental health experiences [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], including EDs [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Platforms (e.g., Meta and Twitter/X in Western countries, and Weibo in China) have provided open and (semi-)anonymous spaces for people to share their daily thoughts and feelings, seek information, and discuss topics through various media (e.g., text, pictures, videos; [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. These digital traces have been shown to reveal novel behavioral patterns and can complement epidemiological and clinical approaches [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. In China, Weibo functions much like Twitter/X, which allows users to rapidly exchange information and form interactive social networks. Moreover, Chinese natural language processing (NLP) presents unique challenges, including word segmentation and idiomatic expressions [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Such linguistic and cultural features necessitate context-specific approaches rather than extrapolating from English-language research. Thus, rigorous study of Chinese platforms like Weibo is essential to uncover more culturally specific signals of EDs at the population scale.\u003c/p\u003e\u003cp\u003eMoreover, because people in ED communities often share their thoughts, behaviors, and daily lives online [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], the large volumes of user-generated content on social media provide a strong basis for large-scale analysis in digital health [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Although Chinese datasets have linguistic and cultural features that differ from those of English, powerful toolkits, such as Machine Learning (ML), can still extract key signals useful for detecting risk factors and behavioral patterns [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. In particular, advanced deep-learning models can accurately identify users showing ED-related symptoms from social media posts [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. These methods help narrow broad user pools to specific at-risk groups, enabling closer examination of distinct patterns.\u003c/p\u003e\u003cp\u003eED-related topics are crucial to understanding the complexities of individuals\u0026rsquo; mental health conditions. A retrospective literature review examined key trends in ED topics, including treatment, symptoms, and risk factors, from 1990 to 2021, emphasizing the importance of ongoing investigation in this field [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Moreover, researchers are also beginning to utilize social media data and big data analysis to enhance these insights [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. For example, Wang and colleagues focused on detecting and characterizing ED communities on Twitter, which provided insights into community structures and interactions among ED individuals [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eMachine learning can also further accelerate topic discovery at scale by minimizing the need for manual labeling. For instance, algorithms such as Latent Dirichlet Allocation (LDA) automatically extract latent themes related to social support and ED-specific content across large datasets, eliminating the need for intensive manual categorization that was previously required [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Similarly, Zhou and colleagues used the Correlation Explanation (CorEx) models to identify and cluster topics related to ED symptoms (e.g., body image, the consequences of EDs; [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]). Together, these methods demonstrate the potential of social media data to enhance prevention and treatment efforts in the public health sector. However, most existing work focuses on English-language datasets and evaluates a limited set of models. This leaves a gap in the understanding of culture-specific risk factors associated with EDs.\u003c/p\u003e\u003cp\u003eOur study extends Zhou et al.\u0026rsquo;s method on Twitter data by adapting their ML classifiers and topic modeling methods to Weibo, thereby addressing the specific cultural and linguistic differences of Chinese-language social media content. Although Zhou et al. demonstrated the feasibility of English datasets, it remains unclear whether they apply equally well to other unique content [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Compared with Twitter, Weibo differs in most topic distributions and dynamics of social attention, as the entertainment context dominates the topics, and hashtag use also follows Weibo-specific conventions [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. In this environment, body-image norms and shared dieting or eating habits can spread rapidly, often via celebrities, influencers, and trending tags, which might shape public perceptions and potentially amplify the risk of EDs. Additionally, Chinese natural language processing tasks require addressing word segmentation and Part-of-Speech tagging issues, which introduce challenges that are less prominent in English [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Moreover, ED-related topics in East Asian contexts are distinct from those in Western settings. For example, studies have found that Chinese young women reported higher levels of baseline stigmatization compared with Australian students (Yan et al., 2018). Additionally, research has demonstrated that East Asians are more likely to communicate negative experiences through somatic expression than Westerners [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Therefore, these specifications in cultural-linguistic features and platform-specific norms underscore the significance of examining ED-related content patterns in Chinese social media.\u003c/p\u003e\u003cp\u003eThus, we utilize data from Weibo to develop machine-learning techniques for exploring ED-related content in a culturally specific Chinese context. Specifically, our research addresses two key questions. The first step is to systematically assess the effectiveness of five distinct statistical and computational methods (e.g., Convolutional Neural Networks [CNNs], Random Forests [RFs], Extreme Gradient Boost [XGBoost], Support Vector Machines [SVMs], and Logistic Regression [LR]) in identifying ED-related content. The second research question aims at exploring and evaluating ED-related topics extracted from Weibo posts via Latent Dirichlet Allocation (LDA) so that risk factors or behavioural patterns associated with ED symptoms can be further revealed.\u003c/p\u003e\u003cp\u003eBy combining supervised classification with unsupervised topic modeling, we aim to demonstrate a scalable framework for ED surveillance in a non-English context. This approach not only advances methodological innovation but also provides culturally grounded insights into disordered eating, with potential applications for early detection and intervention in China.\u003c/p\u003e"},{"header":"2. Method","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Study Design\u003c/h2\u003e\u003cp\u003eOur study methodology follows a two-stage pipeline shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. In the first stage, we collected an initial dataset of 6,589 Weibo posts containing keywords related to EDs via the Weibo Application Programming Interface (API). These posts are manually annotated into three categories (i.e., ED-laypeople, ED-promotion and education, and ED-irrelevant), which yields 3,469 posts in the ED-laypeople subset.\u003c/p\u003e\u003cp\u003eUsing these 3,469 posts as our training dataset, we developed and compared several traditional ML-classifiers alongside a deep learning (DL) model. The power of ML is utilized to identify ED-related posts and analyze ED-related topics within them, resulting in a twofold development. All models had two main tasks: first, to filter out ED-irrelevant posts and then to sort the remaining posts into either ED-promotion and education or ED-laypeople. Second, to evaluate the effectiveness of the classifiers, we identified the user IDs associated with the 3,469 laypeople posts and randomly sampled 20,000 additional posts from these same users as our testing datasets. Afterwards, the best-performing classifier retrieved 8,665 new laypeople subsets.\u003c/p\u003e\u003cp\u003eIn the second stage, we applied LDA topic modelling to both the manually labelled ED-laypeople subset (N\u0026thinsp;=\u0026thinsp;3,469) and the classifier-identified subset (N\u0026thinsp;=\u0026thinsp;8,665). The outcome of the topic modelling was scrutinized to assess how well it assisted in generating ED-relevant topics for each post. This was achieved by comparing each topic\u0026rsquo;s theme word to the content of posts under that topic, aiming to validate the relevance and accuracy of the automated topic modelling process.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Weibo Posts Collection\u003c/h2\u003e\u003cp\u003eThis study was conducted in January 2023. We used the Weibo streaming API to search and collect posts relevant to our research. Utilizing a carefully selected list of keywords related to EDs (e.g., \u0026ldquo;厌食症 Anorexia\u0026rdquo;, \u0026ldquo;神经性厌食症 Anorexia Nervosa\u0026rdquo;, \u0026ldquo;神经性贪食 Bulimia Nervosa\u0026rdquo;, \u0026ldquo;贪食症 Bulimia\u0026rdquo;, \u0026ldquo;暴食症 Binge-eating Disorder\u0026rdquo;, \u0026ldquo;暴饮暴食 Binge eating\u0026rdquo;, \u0026ldquo;饮食失调 Eating disorder\u0026rdquo;, \u0026ldquo;狂食症 Compulsive overeating\u0026rdquo;, \u0026ldquo;进食障碍 Eating Disorder\u0026rdquo;, \u0026ldquo;易饿病 Excessive Hunger\u0026rdquo;) in Chinese, we were able to retrieve a comprehensive pool of posts for manual screening on Weibo.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Identification of Target Posts\u003c/h2\u003e\u003cdiv id=\"Sec6\" class=\"Section3\"\u003e\u003ch2\u003e2.3.1 Manually Annotated ED-laypeople Posts and ED-irrelevant Posts\u003c/h2\u003e\u003cp\u003eOur framework for identifying target posts was based on the \u0026ldquo;gold-standard\u0026rdquo; dataset distribution reported by Zhou et al., which annotated posts into three categories: \u003cem\u003eED-irrelevant, ED-promotional and education\u003c/em\u003e, and \u003cem\u003eED-laypeopl\u003c/em\u003ee. Posts that were unrelated to EDs were categorized as ED-irrelevant; posts aimed at promoting products or educating the public about EDs were classified as \u003cem\u003eED-promotional and education\u003c/em\u003e posts; and \u003cem\u003eED-laypeople\u003c/em\u003e posts referred to ED-relevant posts made by individual users. In this study, we focused specifically on \u003cem\u003eED-laypeople\u003c/em\u003e posts [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThree reviewers and two supervisors formed the annotation team for the manual review tasks in the study. All team members were pre-trained on the annotation principles to ensure consistency. Initially, the three reviewers independently annotated the posts. Then, they cross-checked their results to ensure the reliability of the results. Next, any inconsistencies were discussed with the supervisors to reach a final consensus on all posts.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section3\"\u003e\u003ch2\u003e2.3.2 Machine Learning Classifier Development\u003c/h2\u003e\u003cp\u003eTo explore latent patterns within the manually labeled corpus, we employed ML algorithms to design automatic text classifiers. Since these algorithms only process numerical data, we pre-processed the texts. This involved removing elements that were potentially devoid of significant information, such as Chinese stop words (e.g., \u0026ldquo;但\u0026rdquo; [d\u0026agrave;n], translating to \u0026ldquo;but\u0026rdquo;); user mentions (e.g., @username); hyperlinks (e.g., \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://abccc.com\u003c/span\u003e\u003cspan address=\"https://abccc.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e); hashtags (e.g., #gym); and emojis. The refined text was then segmented into individual words using the \u0026lsquo;\u003cem\u003ejieba\u003c/em\u003e\u0026rsquo; Chinese text segmentation toolkit. Subsequently, a comprehensive lexicon was established, assigning a unique numerical identifier to each word. This conversion transformed each post into a numeric vector representation.\u003c/p\u003e\u003cp\u003eOur study centered on the construction of a classifier utilizing four conventional machine learning models, including Logistic Regression (LR), Support Vector Machines (SVMs), Random Forests (RFs), and Gradient Boosting Trees (XGBoost), alongside Convolutional Neural Networks (CNNs), a DL model. To enhance performance and avoid overfitting, we meticulously tuned the hyperparameters of the traditional models by amalgamating grid search with cross-validation.\u003c/p\u003e\u003cp\u003eThe CNNs\u0026rsquo; structure consisted of three layers: an embedding layer, a convolutional block, and an output layer. The embedding layer, pre-populated with terms from both corpora mentioned above, facilitates the initial transformation of text into vector form. The convolutional block sequentially applied 1D convolutions with 64 filters of kernel size 3, followed by the ReLU activation function, max pooling with a kernel size of 2 for dimensionality reduction, and a dropout with a rate of 0.2 to deter overfitting. The last linear output layer was responsible for projecting the flattened features from the CNN onto the spectrum of class categories, yielding logits corresponding to each class.\u003c/p\u003e\u003cp\u003eOur classification framework was designed in two stages. The initial phase entailed differentiating ED-related posts (including ED-layperson discussions and ED-promotion and education posts) from those irrelevant to ED, resulting in a binary classification. The ensuing stage engaged classifiers to distinguish between posts of ED-laypeople and those categorized under ED-promotion and education.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Identifying Topics from Posts of Potential ED Users\u003c/h2\u003e\u003cp\u003eTopic modeling techniques were used to extract topics and explore the mental status and daily concerns of potential ED users. Specifically, we utilized Latent Dirichlet Allocation (LDA; [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]), a probabilistic model designed to identify latent thematic structures within sizable textual datasets. This model assumes that documents are composed of a blend of latent topics, where each topic is defined by its unique word distribution. In practice, LDA interprets each document as a mixture of topics, assigning each word to a topic based on the distribution of words across topics and vice versa.\u003c/p\u003e\u003cp\u003eWe applied the LDA model to two distinct sets of data: (1) the initial dataset's posts that were manually labeled as ED-related discussions by laypeople and (2) specific posts from potential ED users, which were categorized as ED-laypeople discussions by our automated classifier.\u003c/p\u003e\u003cp\u003eDetermining the appropriate number of topics in LDA is a pivotal step. We experimented with different topic quantities, ranging from 4 to 10. While statistical measures such as perplexity and relevance scores offer guidance on the number of topics, they do not always yield thematically rich topics. Therefore, we conducted a series of testimonies and meticulously examined the resultant topics, allowing us to identify the most informative and representative topic number for subsequent in-depth analysis.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Topic evaluation\u003c/h2\u003e\u003cp\u003eOur team was instructed by domain experts to manually identify each topic from the top 30 most salient keywords generated by topic modeling and assign a corresponding key theme to each topic. Next, we then manually reviewed the top 10 most relevant posts associated with these keywords and evaluated whether the post was coherent with the generated topic theme. The relevance scores were defined as the percentages of posts that were coherent with each topic theme.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Post Collection and Manual Identification of ED-related Content\u003c/h2\u003e\u003cp\u003eUsing the keyword search feature via the Weibo API, we collected 6,589 posts. Of these, 3,489 were classified as discussions by ED laypeople, 1,338 as ED promotional and educational materials, and 1,782 as unrelated to EDs. The results of Cohen\u0026rsquo;s Kappa range from .58 to .79, indicating a moderate to substantial agreement among reviewers. The post collections of users whose posts were labeled as ED laypeople are the source of the second dataset, which consists of 20,000 posts randomly selected from these collections.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Machine Learning Classifier Development\u003c/h2\u003e\u003cp\u003eOur analysis encompassed five machine learning algorithms: CNNs, RFs, XGBoost, SVMs, and LR, mentioned above. The manually labeled dataset was randomly divided into a training set, encompassing 69.98% of the posts (4,611 out of 6,589), and a validation set, containing the remaining 30.02% (1,978 out of 6,589). Stratified sampling ensured equivalent distribution across both sets relative to the original dataset. We designed a two-tiered classification system to separate ED-irrelevant posts and to differentiate between ED-promotional and ED-laypeople discussions. Preliminary results indicate the classifiers' performances as shown in the subsequent Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePerformance comparison of classifiers under different thresholds\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\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\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eClassifier\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eF1-score\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eF1-score\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eED-irrelevant versus other 2 labels (Threshold 0.5)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e\u003cp\u003eED-irrelevant (535)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e\u003cp\u003eED-promotional and education\u0026thinsp;+\u0026thinsp;ED-laypeople (1443)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.68\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.69\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.68\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.89\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.88\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.88\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.83\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.55\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.66\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.85\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.96\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.81\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.51\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.63\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.84\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.96\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.68\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.36\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.47\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.94\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.86\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.57\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.46\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.51\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.82\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.87\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.84\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eED-irrelevant versus other 2 labels (Threshold 0.9)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e\u003cp\u003eED-irrelevant (535)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e\u003cp\u003eED-promotional and education\u0026thinsp;+\u0026thinsp;ED-laypeople (1443)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.75\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.69\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.85\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.87\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.35\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.96\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.51\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.96\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.34\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.5\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.29\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.98\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.45\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.96\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.21\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.37\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.83\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.51\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.89\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.48\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.62\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.31\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.92\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.46\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.89\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.25\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.39\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eED-promotional and education versus ED-laypeople\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e\u003cp\u003eED-promotional and education (402)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e\u003cp\u003eED-laypeople (1041)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.99\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.91\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.95\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.97\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.98\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.95\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.92\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.93\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.97\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.98\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.97\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.91\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.87\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.89\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.95\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.97\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.96\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.87\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.75\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.91\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.96\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.93\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.88\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.88\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.96\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.92\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 the initial stage, we examined two methodologies to distinguish ED-irrelevant posts from the other two groups. For any post input, all the models were regarded as statistical models that output the probability of the input belonging to ED-irrelevant posts and the probability of the opposite. If the traditional 0.5 threshold is set, events with higher probability are more likely to occur, so the post should belong to the higher probability group. By setting the conventional threshold at 0.5, our models exhibited a commendable average F1-score of 0.87 but a modest average recall of 0.51 for the ED-irrelevant category, indicating that although the classifier successfully found ED-relevant posts, the posts predicted as no ED-irrelevant posts would be mixed with a considerable number of ED-irrelevant posts. Therefore, we set the threshold at a high value of 0.9, meaning the classifier considers the post ED-irrelevant only if the output probability exceeds 0.9. The adjustment of the threshold diminished the F1-scores for the traditional models; however, the CNN sustained its high F1-score (0.87) and enhanced the recall for ED-irrelevant posts to 0.75.\u003c/p\u003e\u003cp\u003eDuring the second phase, the CNNs surpassed their traditional counterparts with an F1-score of 0.98. Consequently, we selected CNNs for automated text classification, which identified 8,665 posts as ED laypeople discussions within the second dataset of 20,000 posts of potential ED users.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e3.3 Topic Modeling and Evaluation\u003c/h2\u003e\u003cp\u003eApplying the LDA model to the aforementioned corpora, 3,489 manually tagged ED laypeople posts and 8,665 CNN-classified ED laypeople posts, enabled topic extraction. Following extensive experimentation and analysis, we determined that five topics are optimal for clustering ED-related symptoms into corresponding ED topics. The emblematic themes associated with each topic are presented subsequently.\u003c/p\u003e\u003cp\u003eThe five topic themes that we extracted based on our manual labeling training data were (1) restrictive ED symptomatology \u0026amp; physical distress, (2) binge eating behaviors \u0026amp; body-health concerns, (3) ED episodes, relapse, \u0026amp; coping narratives, (4) emotional venting during ED episodes, and (5) chronic ED patterns \u0026amp; identity impact. The relevance scores for these topics were 90%, 90%, 90%, 80%, \u0026amp; 100%, respectively. According to the intertopic distance map (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), topic theme 1 and topic theme 2 were positioned far apart, with large and non-overlapping bubbles, indicating two highly prevalent but lexically distinct subtopics. In contrast, topics 3, 4, and 5 formed a tight cluster with partial overlaps. This reflected shared topics related to ED episodes but with differentiated focuses. Specifically, topic theme 3 focused on diary-style and relapse-related language; topic theme 4 centered on affect-laden expression; and topic theme 5 emphasized the habitual-related impacts of chronic ED patterns (Table\u0026nbsp;2). Additional figures also showed the top-30 most relevant terms for each topic (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eTable 2: Five topic themes and their 10 most relevant contents\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"554\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" style=\"width: 554px;\"\u003e\n \u003cp\u003eTopic Theme 1: restrictive ED symptomatology \u0026amp; physical distress\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003eContent keywords\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eKeywords translation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003eRelevance\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e饿得想吐,\u0026nbsp;单纯想吐,\u0026nbsp;厌食症倾向\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003enausea due to hunger/ plain nausea, anorexia tendency*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e头晕眼花,\u0026nbsp;喘大气,\u0026nbsp;不想吃饭\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003edizzy, short of breath, no appetite\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e减肥,一天不吃还没感觉, \u0026nbsp;厌食症倾向\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ediet, no hunger after a day without food, anorexia tendency\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e饿/涨得睡不着,\u0026nbsp;暴饮暴食，\u0026nbsp;想吐\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003einsomnia due to hunger/bloating; binge eating, want to vomit\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e新冠后遗症,\u0026nbsp;厌食症,\u0026nbsp;恶心,\u0026nbsp;瘦八斤\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003esequela of covid-19, anorexia*, nausea, lost 4kg\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e感觉得了厌食症,\u0026nbsp;吃饭像完成任务\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003efeel I have anorexia, eating like a task\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e恶心,吃不下,\u0026nbsp;新冠后遗症，厌食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003enausea, no appetite, sequela of covid-19, anorexia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e巨能吃,暴饮暴食\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ehuge appetite, binge eating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e吃一两口就饱,\u0026nbsp;吐,\u0026nbsp;怀疑有厌食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003efull after several bites, vomit, suspect of anorexia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e没饥饿感,\u0026nbsp;是不是有厌食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eNo hunger, suspect of anorexia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" style=\"width: 198px;\"\u003e\n \u003cp\u003eRelevance Scores\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e90%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" style=\"width: 554px;\"\u003e\n \u003cp\u003e* Anorexia tendency and anorexia means low appetite in China, incidating restrictive ED symptomology. People may use sentences like \u0026quot;I feels like I have gotten the anorexia that I have no appetite these days.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" style=\"width: 554px;\"\u003e\n \u003cp\u003eTopic Theme 2: binge eating behaviors \u0026amp; body-health concerns\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003eContent keywords\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eKeywords translation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003eRelevance\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e暴饮暴食然后减肥,\u0026nbsp;熬夜\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebinge eating then diet, stay up\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e暴饮暴食两周多,\u0026nbsp;体重高\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebinge eating over 2 weeks, high weight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e天天暴饮暴食\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebinge eating daily\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e暴饮暴食\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebinge eating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e暴饮暴食\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebinge eating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e严重的抑郁症、厌食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003esevere depression, anorexia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e夜间狂食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003enight-eating syndrome\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e暴饮暴食诱发寻麻疹\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebinge eating triggered hives\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e免疫力差，躯体化反应，进食障碍\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003epoor immune system, somatization reaction, ED\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e暴食症,执着进食\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebinge eating disorder, eating obsessively\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" style=\"width: 198px;\"\u003e\n \u003cp\u003eRelevance Scores\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e90%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" style=\"width: 554px;\"\u003e\n \u003cp\u003eTopic Theme 3: ED episodes, relapse, \u0026amp; coping narratives\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003eContent keywords\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eKeywords translation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003eRelevance\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e***神经性贪食\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ef*** bulimia nervosa\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e打卡,\u0026nbsp;暴食症\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003emeal log, binge eating disorders\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e暴食症的康复日记,进食障碍\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebulimia recovery diary, ED\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e贪食症复发,心情很受影响\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003erelapse of binge eating, mood affecting\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e暴食症,\u0026nbsp;考试\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebinge eating disorders, tests\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e稳定暴食欲望\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003estabilize cravings\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e贪食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebulimia nervosa\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e今天没暴食,\u0026nbsp;不然又要后悔\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eno binge today, no regret\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e最近暴食症很严重\u0026hellip;狂塞,\u0026nbsp;塞满胃部,\u0026nbsp;放弃思考\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003esevere binge eating recently, stuffing stomach until full, give up thinking\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e暴食症的康复日记,气色不好,\u0026nbsp;停止暴食,\u0026nbsp;爱自己\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebinge eating recovery dairy, pale, stop binge, love yourself\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" style=\"width: 198px;\"\u003e\n \u003cp\u003eRelevance Scores\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e90%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" style=\"width: 554px;\"\u003e\n \u003cp\u003eTopic Theme 4: emotional venting during ED episodes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003eContent keywords\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eKeywords translation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003eRelevance\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e厌食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eanorexia nervosa\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e像是得了厌食症,\u0026nbsp;生无可恋\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003elike anorexia nervosa, nothing to live for\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e厌食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eanorexia nervosa\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e食べ過ぎて気持ち悪いっ,狂食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eI feel sick from eating too much, eating syndrome\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e狂食症的反例\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eCounterexample of binge eating disorders\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e厌食症,\u0026nbsp;狂食症,\u0026nbsp;心碎\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eAnorexia nervosa, eating syndrome, heart breaking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e无语,\u0026nbsp;狂食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eI can\u0026rsquo;t even, eating syndrome\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e狂食症, ~!\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eeating syndrome with excited tones\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e狂食症, !!\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eeating syndrome with excited tones\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e狂食症,\u0026nbsp;真坑爹\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eeating syndrome, so unfair\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" style=\"width: 198px;\"\u003e\n \u003cp\u003eRelevance Scores\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e70%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" style=\"width: 554px;\"\u003e\n \u003cp\u003eTopic Theme 5: chronic ED patterns \u0026amp; identity impact\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003eContent keywords\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eKeywords translation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003eRelevance\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e兴趣爱好,\u0026nbsp;乱消费,\u0026nbsp;暴饮暴食,\u0026nbsp;一整天不出门\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ehobbies, overspending, binge eating, staying in bed all day\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e熬夜喝酒,\u0026nbsp;暴饮暴食,\u0026nbsp;早点猝死\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003e\u0026nbsp;Staying up late drinking, binge eating, planning to drop dead early\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e暴饮暴食,\u0026nbsp;排便\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ebinge eating, bowel movement\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e狂食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eeating syndrome\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e暴饮暴食株,\u0026nbsp;撑的要死\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003estrain, binge eating, feel unbearably stuffed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e兴趣爱好,\u0026nbsp;乱消费,\u0026nbsp;暴饮暴食,\u0026nbsp;一整天不出门\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ehobbies, overspending, binge eating, staying in bed all day\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e兴趣爱好,\u0026nbsp;乱消费,\u0026nbsp;暴饮暴食,\u0026nbsp;一整天不出门\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003ehobbies, overspending, binge eating, staying in bed all day\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e鸭梨山大,\u0026nbsp;狂食症\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eoverwhelming stress, eating syndrome\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e狂吃嗜睡\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003eeating syndrome, oversleeping\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 46px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 152px;\"\u003e\n \u003cp\u003e不快乐,\u0026nbsp;暴饮暴食\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003efeeling down, binge eating\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" style=\"width: 198px;\"\u003e\n \u003cp\u003eRelevance Scores\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 217px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003e100%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\u003cp\u003eIn addition to the five topic themes described above, our analysis also revealed several cross-cutting patterns in the post contents. Many posts linked ED symptoms with comorbid conditions (e.g., sleeping difficulties, anxiety, endocrine dyscrasia) and with bodily sensations (e.g., lowered immune function, urticaria, somatization). Furthermore, given the specific years represented in our dataset, a notable subset of posts referenced the sequelae of COVID-19.\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThis study developed a two-tier text classification and topic modeling pipeline to identify and characterize ED-related social media posts from laypeople. In Stage 1, the CNN maintained a strong F1-score (0.87). In Stage 2, the CNN outperformed traditional models and was used to classify 20,000 additional posts, subsequently identifying 8,665 posts from ED laypeople. LDA on the combined corpora yielded five interpretable topic themes related to EDs.\u003c/p\u003e\u003cp\u003eOur CNN accurately classified social-media posts, showing that deep models learn predictive local phrase patterns that simpler baselines miss. This finding is consistent with reports that CNNs and other deep neural models perform strongly in text-based mental health detection [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. In parallel, our topic modeling yielded coherent, clinically interpretable themes from unselected posts, adding information that complements the supervised classifier. Prior studies also show that LDA and related models reveal latent concerns, track shifts in discourse, and describe support dynamics in online mental health communities [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe distance between bubbles of Topic themes 1 and 2 suggests that restriction-focused ED symptomology (e.g., \u0026ldquo;no appetite\u0026rdquo;, \u0026ldquo;anorexia tendency\u0026rdquo;) is linguistically apart from binge-eating-focused ED symptoms (e.g., \u0026ldquo;binge eating\u0026rdquo;, weight complaints). The clustering of Topic themes 3, 4, and 5 indicates that laypeople often narrate their ED episodes alongside emotion dysregulation and personal consequences (e.g., staying overnight, overwhelming stress), such that episodic descriptions of EDs and coping strategies (e.g., recovery diaries, \u0026ldquo;staying in bed all day\u0026rdquo;) appear together. These findings align with the literature, which suggests that emotional dysregulation (e.g., emotional nonacceptance, emotional distress, poor impulse control, deficits in emotion awareness) is a central feature across various EDs [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Meanwhile, the presence of diary-style self-monitoring and logs aligns with the CBT-E practice, which encourages the use of a real-time self-monitoring mechanism to modify eating behaviors. In this setting, users can anonymously use social media to track their eating behaviors and support their self-regulation, while often seeking community as well, without disclosing their formal diagnosis or treatment. These five topic themes suggest a need for interventions that not only address cognitive reframing but also provide psychoeducation related to body-focused management (i.e., understanding hunger cues and techniques to soothe nausea), connect individuals to chronic recovery resources (i.e., routine-based self-care and narrative-based techniques), and ensure sustained long-term support.\u003c/p\u003e\u003cp\u003eOur findings are based on ED-related content in user-generated posts and are therefore not suitable for clinical settings. However, as social media offers a medium to reach broader public audiences, including individuals who may conceal or underreport maladaptive eating behaviors, these data could help characterize behavioral patterns in those lay populations and inform targeted psychoeducation courses, interventions, and prevention programs. Additionally, using a stringent 0.9 threshold for the ED-irrelevant class trades precision for recall to balance the results, which may potentially place borderline and/or ambiguous posts in the ED-relevant stream (i.e., ED-promotional and educational content, ED-laypeople). Moreover, although overall relevance scores are high, the topic coherence measure remains partly subjective, which prevents interpretability and leaves room for refinement [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. We could incorporate human-in-the-loop to adjust for those borderline cases to improve reliability in our future work. Further gains may be achieved by using more advanced models. Recurrent models such as LSTM can improve classification accuracy [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], and topic models like BERTopic [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e] can better capture latent patterns in text. At the same time, social-media posts, especially on platforms such as Weibo, often contain noise, including slang, idioms, and ambiguous internet expressions. More careful preprocessing and cleaning, along with the use of higher-quality text sources, when possible, are therefore crucial for obtaining reliable and meaningful results.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eOur study provides actionable insights that CNN can reliably classify large volumes of social media posts and combine with LDA to cluster coherent themes that point to intervention targets. Practically, this approach can help public-health teams prioritize designs targeted to psychoeducational content for lay communities who may not disclose clinical diagnoses. For example, this offers a novel direction for implementing interventions focused on physical sensation, such as hunger cues, in EDs, and initiating long-term recovery management (i.e., routine-based recovery supports, narrative-based self-monitoring tools). For future work, we recommend incorporating human-in-the-loop review for borderline cases, utilizing more advanced topic and text models, refining platform-specific preprocessing, and validating findings against clinical samples to enhance reliability and inform the design of intervention trials more effectively.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eAN\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eAnorexia Nervosa\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eBED\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eBinge Eating Disorder\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eBN\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eBulimia Nervosa\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eCNN\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eConvolutional Neural Network\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eDL\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eDeep Learning\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eED\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eEating Disorder\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eLDA\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eLatent Dirichlet Allocation\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eLR\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eLogistic Regression\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eML\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eMachine Learning\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eNLP\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eNatural Language Processing\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eRF\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eSVM\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eSupport Vector Machine\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eXGBoost\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eExtreme Gradient Boosting\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eTable\u0026nbsp;2\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eFive topic themes and their 10 most relevant contents\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthical approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eEthical approval\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe project was approved by the Institutional Review Board (Applied Psychology) of The Chinese University of Hong Kong, Shenzhen (Approval No. EF20241018001).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eInformed consent\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eNot Applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data are available from the corresponding author upon request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors(s) declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eJinbo He was supported by the National Natural Science Foundation of China (Grant Number 72204208). Feng Ji was supported by the Connaught Fund (Grant Number 520245) and Social Sciences and Humanities Research Council (SSHRC) of Canada (Grant Number 215119, CRC-2024-00169).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eYuchen Zhang: Investigation, Project administration, Writing \u0026ndash; original draft, Writing \u0026ndash; review \u0026amp; editing; Nanyu Luo: Formal analysis, Writing \u0026ndash; original draft, Writing \u0026ndash; review \u0026amp; editing; Xiaoya Zhang: Writing \u0026ndash; review \u0026amp; editing, Methodology; Feng Ji: Conceptualization, Supervision, Funding acquisition, Project administration, and Writing \u0026ndash; review \u0026amp; editing; Jinbo He: Conceptualization, Supervision, Funding acquisition, Project administration, and Writing \u0026ndash; review \u0026amp; editing.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration of generative AI in scientific writing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDuring the preparation of this work, the authors utilized GPT-5 to enhance the language of the manuscript. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot Applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eKlump KL, et al. Academy for eating disorders position paper: eating disorders are serious mental illnesses. Int J Eat Disord. 2009;42(2). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://doi.org/10.1002/eat.20589\u003c/span\u003e\u003cspan address=\"10.1002/eat.20589\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKrug I et al. \u003cem\u003eA meta-analysis of mortality rates in eating disorders: an update of the literature from 2010 to 2024.\u003c/em\u003e Clinical psychology review, 2025: p. 102547. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cpr.2025.102547\u003c/span\u003e\u003cspan address=\"10.1016/j.cpr.2025.102547\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSmink FR, Van Hoeken D, Hoek HW. Epidemiology of eating disorders: incidence, prevalence and mortality rates. Curr psychiatry Rep. 2012;14(4):406\u0026ndash;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11920-012-0282-y\u003c/span\u003e\u003cspan address=\"10.1007/s11920-012-0282-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCarlson JL, Lemly DC. Medical Considerations and Consequences of Eating Disorders. Focus. 2024;22(3):301\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1176/appi.focus.20230042\u003c/span\u003e\u003cspan address=\"10.1176/appi.focus.20230042\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMitchell JE. Medical comorbidity and medical complications associated with binge-eating disorder. Int J Eat Disord. 2016;49(3):319\u0026ndash;23. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/eat.22452\u003c/span\u003e\u003cspan address=\"10.1002/eat.22452\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKeski-Rahkonen A, Mustelin L. Epidemiology of eating disorders in Europe: prevalence, incidence, comorbidity, course, consequences, and risk factors. Curr Opin Psychiatry. 2016;29(6):340\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://doi.org/10.1097/YCO.0000000000000278\u003c/span\u003e\u003cspan address=\"10.1097/YCO.0000000000000278\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChavez M, Insel TR. Eating disorders: National Institute of Mental Health's perspective. Am Psychol. 2007;62(3):159. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1037/0003-066X.62.3.159\u003c/span\u003e\u003cspan address=\"10.1037/0003-066X.62.3.159\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi Z, et al. Burden of eating disorders in China, 1990\u0026ndash;2019: an updated systematic analysis of the global burden of disease study 2019. Front Psychiatry. 2021;12:632418. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpsyt.2021.632418\u003c/span\u003e\u003cspan address=\"10.3389/fpsyt.2021.632418\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eQian J et al. \u003cem\u003eAn update on the prevalence of eating disorders in the general population: a systematic review and meta-analysis.\u003c/em\u003e Eating and Weight Disorders-Studies on Anorexia, Bulimia and Obesity, 2022. 27(2): pp. 415\u0026ndash;28. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s40519-021-01162-z\u003c/span\u003e\u003cspan address=\"10.1007/s40519-021-01162-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou S, et al. Exploring eating disorder topics on Twitter: machine learning approach. JMIR Med Inf. 2020;8(10):e18273. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://doi.org/10.2196/18273\u003c/span\u003e\u003cspan address=\"10.2196/18273\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHoward LM, et al. \u003cem\u003eThe role of denial in eating disorder development, assessment, and treatment\u003c/em\u003e, in \u003cem\u003eEating Disorders\u003c/em\u003e. Springer; 2023. pp. 367\u0026ndash;83.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWu J, et al. Secular trends in the incidence of eating disorders in China from 1990 to 2017: a joinpoint and age\u0026ndash;period\u0026ndash;cohort analysis. Psychol Med. 2022;52(5):946\u0026ndash;56. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1017/S0033291720002706\u003c/span\u003e\u003cspan address=\"10.1017/S0033291720002706\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDe Choudhury M et al. \u003cem\u003ePredicting depression via social media\u003c/em\u003e. in \u003cem\u003eProceedings of the international AAAI conference on web and social media\u003c/em\u003e. 2013. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1609/icwsm.v7i1.14432\u003c/span\u003e\u003cspan address=\"10.1609/icwsm.v7i1.14432\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChancellor S et al. \u003cem\u003eQuantifying and predicting mental illness severity in online pro-eating disorder communities\u003c/em\u003e. in \u003cem\u003eProceedings of the 19th ACM conference on computer-supported cooperative work \u0026amp; social computing\u003c/em\u003e. 2016. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/2818048.2819973\u003c/span\u003e\u003cspan address=\"10.1145/2818048.2819973\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDerenne J, Beresin E. Body image, media, and eating disorders\u0026mdash;a 10-year update. Acad Psychiatry. 2018;42(1):129\u0026ndash;34. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s40596-017-0832-z\u003c/span\u003e\u003cspan address=\"10.1007/s40596-017-0832-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang X, et al. Depression risk prediction for chinese microblogs via deep-learning methods: Content analysis. JMIR Med Inf. 2020;8(7):e17958. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://doi.org/10.2196/17958\u003c/span\u003e\u003cspan address=\"10.2196/17958\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSproat R, Emerson T. \u003cem\u003eThe first international Chinese word segmentation bakeoff\u003c/em\u003e. in \u003cem\u003eProceedings of the second SIGHAN workshop on Chinese language processing\u003c/em\u003e. 2003.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZheng C, Huang M, Sun A. \u003cem\u003eChID: A large-scale Chinese IDiom dataset for cloze test.\u003c/em\u003e arXiv preprint arXiv:1906.01265, 2019.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePark E, Kim W-H, Kim S-B. \u003cem\u003eWhat topics do members of the eating disorder online community discuss and empathize with? an application of big data analytics\u003c/em\u003e. in \u003cem\u003eHealthcare\u003c/em\u003e. 2022. MDPI. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/healthcare10050928\u003c/span\u003e\u003cspan address=\"10.3390/healthcare10050928\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang T et al. \u003cem\u003eDetecting and characterizing eating-disorder communities on social media\u003c/em\u003e. in \u003cem\u003eProceedings of the Tenth ACM International conference on web search and data mining\u003c/em\u003e. 2017. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/3018661.3018706\u003c/span\u003e\u003cspan address=\"10.1145/3018661.3018706\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMathews SC, et al. Digital health: a path to validation. NPJ Digit Med. 2019;2(1):38. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41746-019-0111-3\u003c/span\u003e\u003cspan address=\"10.1038/s41746-019-0111-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGuan L, et al. Identifying Chinese microblog users with high suicide probability using internet-based profile and linguistic features: classification model. JMIR mental health. 2015;2(2):e4227.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang X et al. \u003cem\u003eDetecting suicidal ideation in Chinese microblogs with psychological lexicons\u003c/em\u003e. in. 2014 \u003cem\u003eIEEE 11th Intl Conf on Ubiquitous Intelligence and Computing and 2014 IEEE 11th Intl Conf on Autonomic and Trusted Computing and 2014 IEEE 14th Intl Conf on Scalable Computing and Communications and Its Associated Workshops\u003c/em\u003e. 2014. IEEE. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://doi.org/10.1109/UIC-ATC-ScalCom.2014.48\u003c/span\u003e\u003cspan address=\"10.1109/UIC-ATC-ScalCom.2014.48\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePark E, Kim W-H. A retrospective literature review of eating disorder research (1990\u0026ndash;2021): application of bibliometrics and topical trends. Int J Environ Res Public Health. 2022;19(13):7710. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ijerph19137710\u003c/span\u003e\u003cspan address=\"10.3390/ijerph19137710\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMoessner M, et al. Analyzing big data in social media: Text and network analyses of an eating disorder forum. Int J Eat Disord. 2018;51(7):656\u0026ndash;67. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/eat.22878\u003c/span\u003e\u003cspan address=\"10.1002/eat.22878\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang Q, Gon\u0026ccedil;alves B. \u003cem\u003eTopical differences between chinese language Twitter and sina weibo\u003c/em\u003e. in \u003cem\u003eProceedings of the 25th International Conference Companion on World Wide Web\u003c/em\u003e. 2016. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/2872518.2890562\u003c/span\u003e\u003cspan address=\"10.1145/2872518.2890562\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu Q, Wang Z. A Data-Driven Model for Automated Chinese Word Segmentation and POS Tagging. Comput Intell Neurosci. 2022;2022(1):7622392. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2022/7622392\u003c/span\u003e\u003cspan address=\"10.1155/2022/7622392\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChoi E, Chentsova-Dutton Y, Parrott WG. The effectiveness of somatization in communicating distress in Korean and American cultural contexts. Front Psychol. 2016;7:383. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpsyg.2016.00383\u003c/span\u003e\u003cspan address=\"10.3389/fpsyg.2016.00383\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRyder AG, et al. The cultural shaping of depression: somatic symptoms in China, psychological symptoms in North America? J Abnorm Psychol. 2008;117(2):300. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://doi.org/10.1037/0021-843X.117.2.300\u003c/span\u003e\u003cspan address=\"10.1037/0021-843X.117.2.300\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBlei DM, Ng AY, Jordan MI. Latent dirichlet allocation. J Mach Learn Res. 2003;3(Jan):993\u0026ndash;1022.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKour H, Gupta MK. An hybrid deep learning approach for depression prediction from user tweets using feature-rich CNN and bi-directional LSTM. Multimedia Tools Appl. 2022;81(17):23649\u0026ndash;85. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11042-022-12648-y\u003c/span\u003e\u003cspan address=\"10.1007/s11042-022-12648-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLow DM, et al. Natural language processing reveals vulnerable mental health support groups and heightened health anxiety on reddit during covid-19: Observational study. J Med Internet Res. 2020;22(10):e22635. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://doi.org/10.2196/22635\u003c/span\u003e\u003cspan address=\"10.2196/22635\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGratz KL, Roemer L. Multidimensional assessment of emotion regulation and dysregulation: Development, factor structure, and initial validation of the difficulties in emotion regulation scale. J Psychopathol Behav Assess. 2004;26(1):41\u0026ndash;54. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1023/B:JOBA.0000007455.08539.94\u003c/span\u003e\u003cspan address=\"10.1023/B:JOBA.0000007455.08539.94\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLavender JM, et al. Dimensions of emotion dysregulation in bulimia nervosa. Eur Eat Disorders Rev. 2014;22(3):212\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://doi.org/10.1016/j.cpr.2015.05.010\u003c/span\u003e\u003cspan address=\"10.1016/j.cpr.2015.05.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eR\u0026ouml;der M, Both A, Hinneburg A. \u003cem\u003eExploring the space of topic coherence measures\u003c/em\u003e. in \u003cem\u003eProceedings of the eighth ACM international conference on Web search and data mining\u003c/em\u003e. 2015. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/2684822.2685324\u003c/span\u003e\u003cspan address=\"10.1145/2684822.2685324\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGrootendorst M. \u003cem\u003eBERTopic: Neural topic modeling with a class-based TF-IDF procedure.\u003c/em\u003e arXiv preprint arXiv:2203.05794, 2022.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"journal-of-eating-disorders","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"joed","sideBox":"Learn more about [Journal of Eating Disorders](http://jeatdisord.biomedcentral.com)","snPcode":"40337","submissionUrl":"https://submission.nature.com/new-submission/40337/3","title":"Journal of Eating Disorders","twitterHandle":"@JEatDisord","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"machine learning, deep learning, eating disorder topics, topic modelling, social media","lastPublishedDoi":"10.21203/rs.3.rs-7852043/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7852043/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003eEating disorders (EDs) are severe psychiatric conditions with high mortality and substantial medical complications. In China, underdiagnosis and low treatment engagement hinder timely intervention. Social media platforms provide a naturalistic lens into ED-related experiences, yet research on Chinese-language data remains scarce. Advances in machine learning (ML) and deep learning (DL) offer new opportunities to identify and characterize such ED discourse, informing the development of scalable detection methods and culturally tailored prevention and intervention strategies in the Chinese context.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eWe collected ED-related posts from Weibo via keyword-based API searches and manually annotated them into three groups: irrelevant, promotional/educational content, and layperson posts. Five ML/DL methods, including Convolutional Neural Networks (CNNs), Random Forests, XGBoost, Support Vector Machines (SVMs), and Logistic Regression, were trained to identify ED-related posts in a two-stage framework: (1) filtering out irrelevant posts and (2) distinguishing promotional/educational posts from layperson posts. Classifier performance was evaluated on additional posts from the same users. Latent Dirichlet Allocation (LDA) was applied to the layperson subset to extract underlying ED-related themes.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eCNN consistently outperformed other models, achieving high F1-scores in both classification stages (0.87 and 0.98, respectively). Topic modelling revealed five themes: restrictive symptomatology and physical distress, binge eating and body-health concerns, relapse and coping narratives, emotional venting, and chronic ED patterns with identity impact.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e\u003cp\u003eThis study demonstrates that CNN-based classification combined with topic modeling provides a scalable framework for detecting ED-related discourse on Chinese social media. Beyond methodological advances in non-English NLP, the findings highlight culturally specific symptom expressions and psychosocial concerns, offering novel insights for public health surveillance. These insights can inform the development of early detection tools and culturally sensitive interventions to address the unmet needs of individuals with EDs in China.\u003c/p\u003e","manuscriptTitle":"Identifying and Characterizing Eating Disorder Discourse on Chinese Social Media: A Machine Learning Approach","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-12 15:14:30","doi":"10.21203/rs.3.rs-7852043/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-05-06T00:34:10+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-26T07:48:29+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"328135064893133709348103417381161184556","date":"2026-03-07T03:43:02+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"234890397874248355065182530492095746407","date":"2025-11-24T02:05:09+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-11-02T19:54:16+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-10-30T10:37:45+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-30T10:35:44+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of Eating Disorders","date":"2025-10-13T19:45:03+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-eating-disorders","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"joed","sideBox":"Learn more about [Journal of Eating Disorders](http://jeatdisord.biomedcentral.com)","snPcode":"40337","submissionUrl":"https://submission.nature.com/new-submission/40337/3","title":"Journal of Eating Disorders","twitterHandle":"@JEatDisord","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"9d916c14-e479-42e7-b5be-b8072617b32e","owner":[],"postedDate":"November 12th, 2025","published":true,"recentEditorialEvents":[{"type":"decision","content":"Revision requested","date":"2026-05-06T00:34:10+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-20T12:38:17+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-12 15:14:30","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7852043","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7852043","identity":"rs-7852043","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}