This paper is in the following e-collection/theme issue:

Public (e)Health, Digital Epidemiology and Public Health Informatics (646) Infoveillance, Infodemiology, Digital Disease Surveillance, Infodemic Management (1015) Infodemiology and Infoveillance (1278) Medicine 2.0: Social Media, Open, Participatory, Collaborative Medicine (1573) Anxiety and Stress Disorders (1013) e-Mental Health and Cyberpsychology (1411)

Published on in Vol 25 (2023)

Preprints (earlier versions) of this paper are available at https://preprints.jmir.org/preprint/45777, first published .
Public Attitudes Toward Anxiety Disorder on Sina Weibo: Content Analysis

Public Attitudes Toward Anxiety Disorder on Sina Weibo: Content Analysis

Public Attitudes Toward Anxiety Disorder on Sina Weibo: Content Analysis

Authors of this article:

Jianghong Zhu1 Author Orcid Image ;   Zepeng Li1 Author Orcid Image ;   Xiu Zhang1 Author Orcid Image ;   Zhenwen Zhang1 Author Orcid Image ;   Bin Hu1 Author Orcid Image
Article Authors Cited by (8) Tweetations (3) Metrics

    Original Paper

    Gansu Provincial Key Laboratory of Wearable Computing, School of Information Science and Engineering, Lanzhou University, Lanzhou, China

    Corresponding Author:

    Bin Hu, PhD

    Gansu Provincial Key Laboratory of Wearable Computing

    School of Information Science and Engineering

    Lanzhou University

    No 222 South Tianshui Road

    Lanzhou, 730000

    China

    Phone: 86 17352120733

    Email: bh@lzu.edu.cn


    Background: Anxiety disorder has become a major clinical and public health problem, causing a significant economic burden worldwide. Public attitudes toward anxiety can impact the psychological state, help-seeking behavior, and social activities of people with anxiety disorder.

    Objective: The purpose of this study was to explore public attitudes toward anxiety disorders and the changing trends of these attitudes by analyzing the posts related to anxiety disorders on Sina Weibo, a Chinese social media platform that has about 582 million users, as well as the psycholinguistic and topical features in the text content of the posts.

    Methods: From April 2018 to March 2022, 325,807 Sina Weibo posts with the keyword “anxiety disorder” were collected and analyzed. First, we analyzed the changing trends in the number and total length of posts every month. Second, a Chinese Linguistic Psychological Text Analysis System (TextMind) was used to analyze the changing trends in the language features of the posts, in which 20 linguistic features were selected and presented. Third, a topic model (biterm topic model) was used for semantic content analysis to identify specific themes in Weibo users’ attitudes toward anxiety.

    Results: The changing trends in the number and the total length of posts indicated that anxiety-related posts significantly increased from April 2018 to March 2022 (R2=0.6512; P<.001 to R2=0.8133; P<.001, respectively) and were greatly impacted by the beginning of a new semester (spring/fall). The analysis of linguistic features showed that the frequency of the cognitive process (R2=0.1782; P=.003), perceptual process (R2=0.1435; P=.008), biological process (R2=0.3225; P<.001), and assent words (R2=0.4412; P<.001) increased significantly over time, while the frequency of the social process words (R2=0.2889; P<.001) decreased significantly, and public anxiety was greatly impacted by the COVID-19 pandemic. Feature correlation analysis showed that the frequencies of words related to work and family are almost negatively correlated with those of other psychological words. Semantic content analysis identified 5 common topical areas: discrimination and stigma, symptoms and physical health, treatment and support, work and social, and family and life. Our results showed that the occurrence probability of the topical area “discrimination and stigma” reached the highest value and averagely accounted for 26.66% in the 4-year period. The occurrence probability of the topical area “family and life” (R2=0.1888; P=.09) decreased over time, while that of the other 4 topical areas increased.

    Conclusions: The findings of our study indicate that public discrimination and stigma against anxiety disorder remain high, particularly in the aspects of self-denial and negative emotions. People with anxiety disorders should receive more social support to reduce the impact of discrimination and stigma.

    J Med Internet Res 2023;25:e45777

    doi:10.2196/45777

    Keywords

    anxiety disorder; linguistic feature; topic model; public attitude; social media 


    Anxiety disorder is a common mental disorder. Anxiety disorders involve repeated episodes of intense anxiety and sudden feelings of fear or terror, which peak in a few minutes (panic attacks). These feelings of anxiety and panic can impact the patient’s normal schooling, work, and life. Anxiety disorders are the most prevalent among adolescents, and according to the statistics of the World Health Organization, 3.6% of the adolescents aged 10-14 years and 4.6% of the adolescents aged 15-19 years worldwide experience anxiety disorders [1]. According to a report on the status of nutrition and chronic diseases in China, the prevalence of anxiety disorders in China reached 4.98% in 2020 [2]. In particular, with the spread of COVID-19, there is a substantial increase in the global cases of anxiety disorders. According to Santomauro et al [3], in a study of 204 countries and regions worldwide, the prevalence of global major depression increased by 26% in 2020.

    Compared to people without anxiety disorders, those with anxiety disorders may have unstable interpersonal relationships, poorer functions, and higher rates of work absenteeism, with significant economic costs and impacts on physical health [4-6]. Moreover, anxiety disorders are associated with a significantly increased mortality risk. Compared with the general population, people with anxiety disorders have 1.4 times increased risk of death from natural causes and 2.5 times increased risk of death from nonnatural causes [7]. The high prevalence, chronicity, and the associated excessive mortality led the World Health Organization to rank anxiety disorders as the ninth leading health-related cause of disability [8], causing a significant economic burden, accounting for 3% of the global burden of disease worldwide and costing about €74.4 billion in 30 European countries [9].

    Several factors prevent people from seeking help for mental illness, including poor quality of services, low levels of health literacy in mental health, and stigma and discrimination. Formal mental health services are not available in many places. Even if they do, they are often unavailable or unaffordable [10]. People often choose to endure mental suffering without relief, rather than risk discrimination and exclusion to access mental health services. The proportion of people with anxiety disorders who received any form of treatment is estimated to be 27.6% worldwide, but only 9.8% received adequate treatment [11]. Shame and stigma are the major obstacles for patients with anxiety disorders to receive an early diagnosis and professional treatment [12-14]. Patients with depression or anxiety disorders (compared to those without psychiatric disorders) are twice as likely to have a stigma, and the association between depression and anxiety is even stronger [15]. In addition, negative public attitudes may adversely impact patients’ social functioning (eg, interpersonal relationships, learning, work abilities) [16].

    Public attitudes toward anxiety disorder can impact the psychology and daily lives of patients with anxiety. Most current research on attitudes toward mental illness use the method of questionnaire surveys [14,15,17]. This is because a questionnaire is an effective way of collecting information in research; other potentially richer measures of attitude, such as qualitative analysis, are often not feasible in large-scale studies due to labor costs [18]. However, the information collected by existing research methods is highly influenced by the subjective will of the investigators. In addition, the existing research methods are still limited in their ability to explore the structure of language expression, which would reflect people’s internal activities. To obtain information on the linguistic expressions of large groups of people, we turn our attention to the information available on social media. As an important medium of mass communication, social media contain a large amount of information that reflects people’s inner activities and emotional states. Users may often present their mental problems or illnesses anonymously through a wide variety of social media or online social health communities [19]. Such an online health community can be a network to express compassion by communicating with people who have similar symptoms [20]. In addition, users often try to obtain health information related to their symptoms on social media in an attempt to diagnose themselves [21,22]. Therefore, the posts on social media provide natural language data on people’s attitudes toward anxiety disorders.

    A large number of studies in recent years have shown that social media data can be used to better understand, identify, and describe mental disorders [23,24] (eg, data from Facebook, Twitter, Instagram, Sina Weibo platforms). Individuals with mental disorders show changes in language and behavior, such as greater negative emotions and heightened self-attentive focus [25-28]. There is a high degree of similarity between patients with different forms of mental distress. Moreover, social media data can be used to assess the anxiety level of users [29,30], identify anxiety disorders [19,31], and assess the anxiety level of the public [32,33], thus enabling the assessment of the anxiety level of a large population of individuals or groups. These assessment methods might make up for the shortcomings of the traditional approach using questionnaires such as the Self-rating Anxiety Scale because it is difficult to ask a large number of individuals to fill in those questionnaires to estimate their anxiety levels separately. In addition, some studies showed that citizens’ perceptions and attitudes toward events can be tracked and discovered by analyzing the content posted on social media platforms [34,35]. Health-related stigmatizing attitudes (eg, depression, suicide) can be identified by analyzing the linguistic features expressed in social media posts [36-39]. Differences in stigmatizing attitudes toward mental health issues can also be reflected in different patterns of language use [40].

    Social media–based research on attitudes toward mental illness is conducted at 2 main levels. The first level of analysis is at the text feature level, using word frequency statistics and affective tendency analysis to infer attitudes. Li et al [37] accurately discovered depression discrimination on social media through linguistic analysis methods. Li et al [40] constructed a model to distinguish schizophrenia-related stigma from depression-related stigma by using some psycholinguistic features automatically extracted from each post. Shen and Rudzicz [19] separated posts on 4 different anxiety-related subreddits from posts on control subreddits with an accuracy of 98%. They used a combination of N-gram language modeling and Linguistic Inquiry and Word Count (LIWC) [41].

    The second level of analysis is topic-based analysis. Compared to text feature analysis, topic-based analysis is a higher and a more meaningful level of research. Topic models are divided into qualitative topic modeling methods and quantitative topic modeling methods. The qualitative topic analysis includes mainly content analysis methods (eg, NVivo software, QSR International), which is a systematic method to summarize the expressed content by inference from the text [42]. Reavley and Pilkington [43] used content analysis methods to find that tweets related to depression and schizophrenia were mostly supportive or neutral, and more than one-third of the tweets reflecting stigmatized attitudes toward depression were mocking or belittling patients. Lachmar et al [44] used the qualitative content analysis method to code tweets containing the tag #My depression looks like# on Twitter in May 2016. The content analysis revealed 7 themes: dysfunctional thoughts, lifestyle challenges, social struggles, hiding behind a mask, apathy and sadness, suicidal thoughts and behaviors, and seeking relief. Although the qualitative analysis method is effective, it usually requires researchers to have clinical expertise, which limits the wide use of the method.

    The main methods of quantitative topic analysis include Latent Dirichlet Allocation (LDA) and its improvement methods. LDA is a statistical model that uses a data-driven unsupervised machine learning process to discover the underlying semantic structure from a series of documents. The latent semantic structure consists of a set of related topics that are identified by the words of co-occurrence. The results of the analysis are presented as word sets of co-occurrence, whose common themes are inferred by the researcher [45,46]. Franz et al [47] used the LDA model to detect content in social media containing self-harming thoughts and self-harming behaviors. Liu and Shi [28] used the LDA model to find that there were 7 main topics discussed by depressed patients on the Sina Weibo platform: negative emotion fluctuation, disease treatment and somatic responses, sleep disorders, sense of worthlessness, suicidal extreme behavior, seeking emotional support, and interpersonal communication. Jo et al [48] used a Structural Topic Model similar to the LDA model [49] to analyze users’ anxiety and worry concerns by analyzing data from 13,148 questions and 29,040 answers related to COVID-19 on Naver, a social networking platform in South Korea, and by using a structured topic model and a method of analyzing the network of words. They found that the long-term situation resulted in a slight increase in the proportion of work-related topics and that people’s anxiety and worry are closely related to physical symptoms and methods of self-protection. Paul and Dredze [50] obtained 13 health-related topics in Twitter data by constructing the Ailment Topic Aspect Model. Sik et al [51] used both quantitative (LDA topic modeling) and qualitative (deep reading) methods to determine the optimal number of topics and their interpretation in depression forums.

    Previous studies on mental disorder attitudes on social media have described attitudes statically at the level of textual features and topic features. There are fewer studies on the dynamics of attitudes over time. Yu et al [38] studied public attitudes toward depression and the changes in these attitudes over time by combining the textual feature level and the topic level. The analysis of linguistic features showed that the frequency of use of emotion, positive emotion, anger, cognition (including the insight subcategory), and conjunctions increases significantly over time. The topic feature results suggested an upward trend in social support for people with depression over time, although there is a persistent stigma attached to depression. Using data from the Reddit platform, Low et al [52] analyzed trends from 90 text-derived features and found that many features including categories such as economic stress, isolation, and home increased significantly during COVID-19, while other categories such as motion decreased significantly. They also combined the LDA model and clustering algorithms and found a significant increase in health anxiety topics in midpandemic posts compared to those in prepandemic posts, with a large number of posts related to health anxiety topics expressing concerns with daily life at home, school, and work. Table 1 summarizes the research objectives and analytical methods of the existing work. For more research on the application of text analysis and natural language processing methods in health care, see [53,54].

    In this study, we explored public attitudes toward anxiety disorders on social media and the changes in the attitudes over time at 2 levels: text features and topic features. Sina Weibo is the most popular social media platform in China, known as the Chinese version of Twitter. With 582 million monthly active users in 2022, Sina Weibo is a valuable source of information for studying attitudes toward anxiety. The purpose of this study was to use this rich data source to study public attitudes toward anxiety and how these attitudes change over time by analyzing the textual features and topic structure of publicly available posts on Chinese social media.

    Table 1. A summary of the research objectives and analytical methods described in studies to date.
    StudyObjective(s)Analytical methods
    Li et al [37]Discover depression discrimination on social mediaText feature analysis
    Li et al [40]Distinguish schizophrenia-related stigma from depression-related stigmaText feature analysis
    Reavley and Pilkington [43]Analyze the emotional polarity in tweets related to depression and schizophreniaQualitative content analysis
    Lachmar et al [44]Reveal the themes on tweets containing the tag #My depression looks like# in May 2016Qualitative content analysis
    Franz et al [47]Detect content in social media containing self-harming thoughts and self-harming behaviorsLatent Dirichlet Allocation model (quantitative topic analysis)
    Liu and Shi [28]Reveal the topics talked about by depressed patients on the Sina Weibo platformLatent Dirichlet Allocation model
    Jo et al [48]Analyze users’ anxiety and worry concerns by analyzing data related to COVID-19 on Naver platformStructural Topic Model and the network of words
    Paul and Dredze [50]Reveal the health-related topics in TwitterAilment Topic Aspect Model
    Sik et al [51]Determine the optimal number of topics and their interpretation in depression forumsLatent Dirichlet Allocation model and qualitative analysis
    Low et al [52]Analyze health anxiety topics in midpandemic posts compared to those in prepandemic postsText feature analysis, Latent Dirichlet Allocation model and clustering algorithm
    Yu et al [38]Analyze public attitudes toward depression and the changes in these attitudes over timeText feature analysis and Latent Dirichlet Allocation model

    Data Collection

    Sina Weibo is one of the most influential Chinese social media platforms. We use the application programming interface of Sina Weibo to obtain posts containing the keyword “焦虑症 (anxiety disorder),” which were posted from April 1, 2018, to March 31, 2022. We collected 403,132 Weibo posts. By a statistical method, we found that these posts were from 295,905 accounts, which means that about 5 in every 10,000 Sina Weibo users were engaged in conversations about anxiety disorders. To better reveal the textual features and topic structure of Weibo posts, before performing the analysis, we preprocessed the original posts according to the following steps:

    1. We deleted the information posted automatically by organizations or Sina Weibo platforms.
    2. We deleted users’ names mentioned by the symbol “@” in the main body of the posts.
    3. We deleted English letters and emoticons.
    4. The keywords “topic” and “super topic” of the posts contain the symbol “#” before and after the body, for example, “#super topic of anxiety#;” so, we used regular expressions to remove all the titles of topics and super topics that appeared in the posts.
    5. We deleted advertising posts whose content recurred many times.
    6. Finally, we deleted posts with very limited information, such as those containing only keywords.

    Through the above preprocessing steps, we finally obtained 325,807 posts with the keyword “anxiety disorder.” To analyze the trend of the number of posts over time, we divided all the posts by month.

    Ethical Considerations

    All the data in this paper were obtained from Sina Weibo’s public data, which protects those who have private profiles from being subject to research studies. Hence, this analysis meets the standards to waive informed consent and similar guidelines [55]. Furthermore, we desensitized the data to protect the privacy of the users. Specifically, we removed all individual information related to the identity of the users.

    Methods of Analysis

    First, the preprocessed posts were divided by month to obtain time series data, and then we analyzed the changes in the number of posts related to anxiety disorders by using a linear regression method (R language; R Core Team and the R Foundation for Statistical Computing). Time series analysis can provide a reasonable mathematical model for the sample data to analyze the information and patterns contained in the sample time series. A time series usually has the following characteristics:

    1. The positions of the data in the data series are determined by the temporal order, and there is a temporal correlation between them.
    2. The values of different points in the data series are somewhat random; therefore, it is difficult to predict completely and accurately from historical observations.
    3. The values with a relatively close time interval have a strong correlation, and this correlation can reflect the system evolution pattern.
    4. Time series data usually tend to have some sort of trend and periodicity, such as seasonality.

    We grouped the text data of the posts by month, from April 1, 2018, to March 31, 2022, and we fit a linear regression by using the number of posts and the total length of posts per month as a function of time. In addition, by analyzing these posts, we found that the number of posts containing the keyword “anxiety disorder” increased dramatically at the beginning of each new term. Approximately 10.67% (34,771/325,807) of the posts contained the keyword “new term begins.” To more clearly analyze the changing trend of the information contained in the posts with the keyword “anxiety disorder,” we selected a subset of the posts that contained no keyword “开学 (new term begins),” which included 291,036 posts (Table S1 in Multimedia Appendix 1).

    Second, we performed textual analysis on the linguistic features of all the posts. The Chinese psychoanalysis tool TextMind [56] was used to conduct statistical analysis on high-frequency words related to the psychological characteristics in posts. TextMind can automatically segment words, classify words, and calculate the frequency of each category of word, using the Chinese lexicon named C-LIWC [57]. C-LIWC is a lexicon containing 32 linguistic features, 32 psychological features (belonging to 6 major categories), and 38 other features such as punctuation marks. In this study, 5 major categories were selected to analyze the trends and relevance characteristics of each category, and further, 11 subcategories of these 5 categories and 4 additional categories were selected to analyze the fine-grained changes (See Table 2 for details).

    Table 2. Categories of the mental characteristics selected in the Linguistic Inquiry and Word Count dictionary.
    CategoryAbbreviationExamples
    Social processSocialinvite, hear, instruct, community, interact, public, culture
    FamilyFamilyson, daughter, husband, parents, uncle, cousin, family
    Affective processAffectserious, excessive, willing, rich, hope, promise, cope
    Positive emotionPosEmoaffectionate, loving, welcome, praise, glorious, interesting, kind
    Negative emotionNegEmoresentful, heartless, failure, worry, trash, protest, abuse
    AnxietyAnx restless, impatient, insomnia, fright, panic, anxiety, nervous
    AngerAngerresentful, angry, enemy, fight, criticize, rage, agitated
    SadnessSadpitiful, disappointed, inferior, sorrowful, suffering, helpless, sad
    Biological processBiosick, fever, healing, tired, pain, numbness, vessels
    BodyBodyfinger, skin, ear, perception, breath, eye, shirt
    HealthHealthlive, care, insomnia, wound, surgery, health, scar
    Cognitive processCogMechaccording, evidence, generally, intend, notice, otherwise, however
    InsightInsightexactly, seem, think, admit, notice, believe, feeling
    ExclusiveExcl regardless, rather, if, otherwise, unless, suppose, however
    Perceptual processPerceptsay, show, watch, listen, feel, touch
    FeelFeelfeel, soft, comfort, fuzzy, sharp, smooth, touch
    (Others) III
    (Others) AssentAssenthah-hah, alright, okay, yes, indeed, sure, clear, good
    (Others) WorkWorkwork, research, postgraduate, study, colleague, interview, unit
    (Others) AchievementAchievesuccess, accomplish, achieve, encourage, reward, plan, effect

    The Chinese psychological analysis system TextMind is a software system developed by the Computational Network Psychology Laboratory of the Institute of Psychology, Chinese Academy of Sciences, for linguistic analysis of Chinese text. TextMind provides users with a full set of analysis solutions from automatic word segmentation in simplified Chinese to linguistic psychological analysis; its lexicon, text, and symbol processing methods are specifically tailored to the simplified Chinese context, and its lexicon classification system is also compatible with LIWC.

    Finally, we used a topic model to analyze the topics of all text data on Sina Weibo for each month. The topic model can generate keywords in each topic and the occurrence probability of each topic appearing in the entire document according to the predetermined number of topics. The LDA and the biterm topic model (BTM) are the classic topic models. The BTM [58] uses biterm (word pairs) to model the text, which can better display the hidden topics in the article; so, the BTM is more suitable for short-text modeling and analysis. Therefore, we chose the BTM to analyze the topics discussed in anxiety-related posts. In the preliminary analysis, the number of topics in BTM was set to 30.


    Trend Analysis of the Number and Total Length of Sina Weibo Posts

    Figure 1 shows the changes in the number and total length of anxiety-related posts from April 2018 to March 2022, as well as the changes in the number of posts excluding the keyword “new term begins.” The number of all anxiety-related posts fluctuated over time and showed a clear upward trend, especially after the outbreak of COVID-19. The upward trend was more clearly illustrated by the changes in the number of posts excluding the keyword “new term begins.” The number of posts increased significantly each year before the start of new terms. However, due to the epidemic, the start of the spring semester in 2020 was postponed to March or April; so, there was no obvious peak. The trend in the total length of anxiety-related posts was almost identical to the trend in the number of posts. Moreover, the increase in the total length of posts was even more remarkable in the early stages of the pandemic. Table 3 shows the results of linear regression. The estimate index in Table 3 shows that regardless of whether there were all anxiety-related posts or anxiety-related posts without the keyword “new term begins,” the changes over time showed a significant increasing trend. The R2 value of the anxiety-related posts without the keyword “new term begins” was 0.8671, indicating that the linear regression model had a better fit.

    In addition, we used the time series decomposition function [59] to obtain the long-term trend, seasonal trend, and random term of the number of anxiety-related posts (see Figure 2). Similar to the results in Table 3, the long-term trend presented an obvious upward trend. In the seasonal trend, there were 2 peaks every year in February and August, but the peak in August was higher. Besides, in China, February and August happen to be the beginning of a new semester (spring/fall) every year, and August is also the beginning of the new school year. Thus, we can conclude that the beginning of new semester has a significant impact on anxiety in Sina Weibo posts.

    ‎
    Figure 1. The trend in the number and total length of Sina Weibo posts.
    Table 3. Linear regression results of the number of Sina Weibo posts.
    ResultsPredictorEstimateSEt test (df)R2P value
    Number of postsTime176.8319.089.267 (46)0.6512<.001
    Number of posts without the keyword “new term begins”Time141.348.15817.32 (46)0.8671<.001
    Total length of postsTime1.27620.09014.16 (46)0.8133<.001
    ‎
    Figure 2. Time series decomposition of the number of Sina Weibo posts.

    Analysis of Linguistic Features

    Analysis of the Features of Major Categories of Words

    As shown in Table 2, 5 major categories of psychological features were chosen: social process (Social), affective process (Affect), biological process (Bio), cognitive process (CogMech), and perceptual process (Percept). We analyzed the occurrence frequency of these 5 categories of words in anxiety-related posts from April 2018 to March 2022 (Table S2 in Multimedia Appendix 1); the results are shown in Figure 3. The words related to CogMech appeared most frequently, followed by those related to Affect, Bio, and Social. The frequency of words related to Percept was the lowest, and the change was not obvious during the 4-year period. The frequency of the words related to CogMech, Affect, and Bio increased during the early stage of COVID-19, from January 2020 to July 2020, while the frequency of the words related to Social decreased. Furthermore, Table 4 shows the linear regression results of the frequency of these 5 major categories of words. The words related to Social showed a downward trend, while the other 4 categories of words showed an upward trend, and the use of words related to CogMech and Bio had the most obvious increasing trend. Moreover, the frequency of these 5 categories of words fluctuated greatly over time; therefore, the R2 values were all small under the linear regression fitting model.

    ‎
    Figure 3. Occurrence frequency of the 5 major linguistic feature words: Social, Affect, Bio, CogMech, and Percept. Affect: affective process; Bio: biological process; CogMech: cognitive process; Percept: perceptual process; Social: social process.
    Table 4. Linear regression results of the occurrence frequency of the 5 major linguistic feature words.

    		PredictorEstimateSEt test (df)R2P value
    SocialaTime–8.907e-052.060e-05–4.323 (46)0.2889.001
    AffectbTime1.224e-052.622e-050.467 (46)0.0047.64
    BiocTime1.452e-043.102e-054.680 (46)0.3225<.001
    CogMechdTime1.967e-046.229e-053.158 (46)0.1782.003
    PercepteTime2.942e-051.060e-052.776 (46)0.1435.008

    aSocial: social process.

    bAffect: affective process.

    cBio: biological process.

    dCogMech: cognitive process.

    ePercept: perceptual process.

    Analysis of the Features of Subcategories of Words

    We selected 11 subcategories out of the 5 major categories (see Table 2) and the other 4 subcategories (I, assent, work, and achieve) to analyze the changing patterns of the various categories of words (Table S2 in Multimedia Appendix 1) and their correlation at a fine-grained level. Table 5 shows the linear regression results of the 15 subcategories of the features. Table 5 demonstrates that the use of words related to negative emotion (NegEmo), anxiety, anger, sadness, body, health, exclusive, feel, I, and assent was on the rise, while the use of words related to family, positive emotion (PosEmo), insight, work, and achievement was on the decline.

    Figure 4 shows the correlation results among the 15 subcategories of features. As shown in the figure, the words related to work and family were almost negatively correlated with other features. Anxiety had the highest correlation with NegEmo, followed by assent with exclusive, NegEmo with health, and health with feel. The features with the highest negative correlation were work and health. In addition to NegEmo, there were insight, health, and feel—all of which were highly correlated with anxiety.

    Table 5. Linear regression results of the 15 subcategories of the 5 features.

    		PredictorEstimateSEt test (df)R2P value
    FamilyTime–2.252e-053.775e-06–5.964 (46)0.4361<.001
    Positive emotionTime–1.623e-051.088e-05–1.491 (46)0.0461.14
    Negative emotionTime3.776e-051.516e-052.491 (46)0.1189.02
    AnxietyTime1.612e-058.281e-061.947 (46)0.0761.06
    AngerTime3.611e-061.331e-062.712 (46)0.1378.009
    SadnessTime7.044e-072.888e-060.244 (46)0.0013.81
    BodyTime8.533e-051.093e-057.809 (46)0.5700<.001
    HealthTime5.264e-051.685e-053.123 (46)0.1750.003
    InsightTime–4.903e-071.204e-05–0.041 (46)3.61e-05.97
    ExclusiveTime7.328e-051.068e-056.861 (46)0.5058<.001
    FeelTime1.281e-054.384e-062.922 (46)0.1566.005
    ITime4.865e-051.254e-053.880 (46)0.2466<.001
    AssentTime7.493e-051.243e-056.027 (46)0.4412<.001
    WorkTime–1.938e-053.645e-05–0.532 (46)0.0061.60
    AchievementTime–3.314e-058.021e-06–4.132 (46)0.2707<.001
    ‎
    Figure 4. Correlation results among the 15 subcategories of features. Achieve: achievement; Anx: anxiety; Excl: exclusive; NegEmo: negative emotion; PosEmo: positive emotion; Sad: sadness.

    Topic Analysis of Sina Weibo Posts

    We performed topic modeling on 325,807 anxiety-related posts from April 2018 to March 2022 by month. Since the setting of the number of topics in the topic model will impact the effect of the model, we compared the results of the models with 20, 30, 50, and 100 topics, and finally, we selected the model with 30 topics for further analysis. In addition, we selected a subset of topics that were easy to infer and mark their meanings. Then, we divided the topics into 5 areas: (1) discrimination and stigma, (2) symptoms and physical health, (3) treatment and support, (4) work and social, and (5) family and life (Table S3 in Multimedia Appendix 1). Each area included at least 3 related topics.

    To make the changing trend of occurrence probability in each area clearer, we calculated the probability of occurrence in each month in each area first and then calculated the average probability of occurrence in each quarter in each area (Figure 5). The results showed that from April 2018 to March 2022, the topical area with the highest probability of occurrence was discrimination and stigma, followed by symptoms and physical health, family and life, and work and social; the lowest was treatment and support. Especially, in the early stage of COVID-19, the occurrence probability of the topical area “discrimination and stigma” reached the highest value, while the occurrence probability of the topical areas “symptoms and physical health” and “family and life” both decreased, reaching almost the lowest value in 4 years. Specifically, the topical area related to discrimination and stigma averagely accounted for 26.66% in the 4-year period and 37.75% in the early stage of COVID-19.

    Table 6 shows the linear regression results of the occurrence probability of the 5 topical areas. It can be seen that “symptoms and physical health” had a significantly positive trend, while “family and life” had a significantly negative trend. In general, combining the results of linguistic feature analysis and topic model, we observed the following results: first, the level of anxiety among Weibo users was significantly affected by the beginning of new terms, which may be due to the high proportion of teenage users. The topic model shows that “work and social” had obvious periodicity, with a high occurrence probability in the first and third quarters of each year and a low occurrence probability in the second and fourth quarters. According to the trend analysis of the number and total length of posts, this area was greatly impacted by topics related to “new term begins.” Second, the public was more anxious around the spring festival. Treatment and support had an obvious upward trend in 2019 and the first quarter of 2021, reaching the minimum in the third quarter. The upward trend was not obvious in early 2020 because of COVID-19, but it still had an upward trend. Third, in the early stage of COVID-19, the public paid more attention to discrimination and stigma, while “symptoms and physical health” and “family and life” showed a significant downward trend. This is consistent with the analysis of the linguistic features in Table 4 that the frequency of CogMech, Affect, and Bio increased, while the frequency of Social decreased. It is worth noting that the public can view anxiety in a positive light, with the topic model showing that family and life is inversely correlated with discrimination and stigma by the changing trend. From the results of the linguistic feature analysis, work and family were negatively correlated with almost all other features. This suggests that people should focus more on family and life to reduce discrimination and stigma.

    ‎
    Figure 5. Average quarterly probability of occurrence of the 5 topics in 325,807 anxiety-related posts from April 2018 to March 2022.
    Table 6. Linear regression results of the probability of occurrence of the topics.

    		PredictorEstimateSEt test (df)R2P value
    Discrimination and stigmaTime4.00e-032.78e-031.440 (14)0.1291.17
    Symptoms and physical healthTime4.36e-031.90e-032.302 (14)0.2746.04
    Treatment and supportTime7.65e-041.42e-030.538 (14)0.02026.60
    Work and socialTime1.48e-032.18e-030.678 (14)0.03176.51
    Family and lifeTime–3.29e-031.82e-03–1.805 (14)0.1888.09

    Changes in Anxiety-Related Sina Weibo Content Over Time

    This study analyzes the changes in anxiety-related post content over time from 3 aspects: post quantity, text characteristics, and topic structure. The results of our study indicated the following:

    1. Between April 2018 and March 2022, there was a clear increase in public interest in anxiety disorders, particularly in terms of symptoms and treatment. According to the trend analysis of the number and total length of posts, whether it is the total number of anxiety-related posts or the number of anxiety-related posts without the keyword “new term begins,” the number of posts increased over time. Moreover, symptoms and physical health showed a significantly positive trend according to the linear regression results of topic frequency in 5 fields.
    2. Public anxiety was greatly impacted by COVID-19, particularly in terms of work and social life. From the analysis of the frequency of anxiety-related posts for the 5 categories, namely, Social, Affect, Bio, CogMech, and Percept, words related to CogMech appeared most frequently, followed by words related to Affect, Bio, and Social. The frequency of words related to Percept was the lowest, and the changes were not obvious during the 4-year period. At the beginning of COVID-19, the linear regression results of the frequency of the 5 major categories of linguistic features showed a downward trend for Social-related words, while the remaining 4 categories showed an upward trend, with the most obvious increase in CogMech-related words. According to the linear regression results of the 15 subcategories, NegEmo, anxiety, anger, sadness, body, health, exclusive, feel, I, and assent showed an upward trend, while family, PosEmo, insight, work, and achievement showed a downward trend.
    3. The public perception of discrimination and stigma against anxiety disorders remained high, especially in terms of self-denial and negative emotions. According to the correlation results among the 15 subcategories in Figure 4, the correlation between anxiety and NegEmo was the highest. Moreover, the most frequent topic in the topic analysis was discrimination and stigma, which was the most discussed topic, especially in the early stage of COVID-19.

    Notably, the topic model also showed that the proportion of posts referring to family and life was lower than that of posts referring to discrimination and stigma. However, according to the average occurrence probability of anxiety-related posts in the topic model, the 2 topics were inversely correlated by trend analysis. In addition, there was a clear seasonal pattern in the topic of symptoms and physical health. Figure 5 shows that the occurrence probability of the topical area “symptoms and health” increased from autumn to winter, and Figure 3 shows that there was a significant increase in the number of the words of Affect and Bio from autumn to winter. This may indicate that people’s anxiety increases from autumn to winter and that the public pays more attention to their health in the winter. This phenomenon was similar to the change in the depression degree among social media users [60]. Previous studies have shown that compared to the control group, users with mental disorders prefer to use first-person pronouns [61]. In this study, we found that the frequency of first-person pronouns in anxiety-related posts remained high and tended to increase over time. This phenomenon suggests that anxiety-related users pay more attention to their CogMech. In addition, our study found a high frequency and a significant increase in the occurrence frequency of the words related to assent in anxiety-related posts. Assent words are usually positively correlated with users’ positive emotions [62,63]. However, in social media, some assent words indicate users’ disdain or helplessness to some extent.

    Limitations

    This study has 3 main limitations. First, social media users cannot represent the entire Chinese population. According to the 50th statistical report on China’s internet development [64], the number of internet users in China reached 1.051 billion by June 2022. Most of these users were adults aged 20-39 years. Concurrently, there were 362 million non–internet users, most of whom lived in rural areas without access to the internet (41.2% of non–internet users live in rural areas). Lack of internet usage skills, literacy restrictions, age factors, and inadequate equipment are the main reasons for non–internet users not to access the internet [64]. Therefore, our results cannot be generalized to non–social media users. The trends of attitude changing over time found in this study may be different from the real trend. Second, the data analyzed in this paper are only for posts containing the keyword anxiety disorder; therefore, some posts that do not contain the keyword but express attitudes toward anxiety disorders are ignored. Third, the results obtained by this simple filtering method are only the overall attitude of Sina Weibo users toward anxiety disorders, lacking fine-grained analysis and without considering the impact of changes in the number of Weibo users. Additionally, there are other social media platforms in China besides Sina Weibo, and the generalization of our findings should be considered cautiously.

    Conclusions and Recommendations

    In this study, the attitudes of Weibo users toward anxiety disorder and the changes in attitudes over time were explored by analyzing the linguistic features and topical structure of anxiety-related posts on Sina Weibo from April 2018 to March 2022. The results of this study showed that public discrimination and stigma against anxiety was high and that public anxiety was greatly impacted by the beginning of a new term and COVID-19. More specifically, the topics related to discrimination and stigma averagely accounted for 26.66% in the 4-year period and 37.75% in the early stage of COVID-19. Approximately 10.67% (34,771/325,807) of the posts contained the keyword “new term begins,” and the number increased dramatically at the beginning of each new term. Fortunately, public awareness of the symptoms and physical health aspects of anxiety disorders has increased significantly over time. Therefore, social media can be used to provide resources and assistance to patients with anxiety disorders and improve social media activities related to discrimination and stigma reduction.

    In the future, we can try to consider English words, emoticons, and other characters in posts to more accurately reveal public attitudes toward anxiety disorder. In addition, by adding more keywords in the process of data collection, more posts on anxiety topics can be acquired and analyzed, which may expand the scope of research and generate more meaningful results.

    Acknowledgments

    This work was supported in part by the Sci-Tech Innovation (STI) 2030-major projects (grant 2021ZD0202000), the National Natural Science Foundation of China (grants 62227807 and 62072219), and the Supercomputing Center of Lanzhou University.

    Conflicts of Interest

    None declared.

    ‎
    Multimedia Appendix 1

    Supplementary data.

    XLSX File (Microsoft Excel File), 28 KB
    1. Mental health of adolescents. World Health Organization.   URL: https://www.who.int/news-room/fact-sheets/detail/adolescent-mental-health [accessed 2022-10-25]
    2. Report on the Nutrition and Chronic Diseases Status of Chinese Residents (2020). Beijing: People's medical Publishing House; 2020.
    3. COVID-19 Mental Disorders Collaborators. Global prevalence and burden of depressive and anxiety disorders in 204 countries and territories in 2020 due to the COVID-19 pandemic. Lancet 2021 Nov 06;398(10312):1700-1712 [FREE Full text] [CrossRef] [Medline]
    4. Plaisier I, Beekman A, de Graaf R, Smit J, van Dyck R, Penninx B. Work functioning in persons with depressive and anxiety disorders: the role of specific psychopathological characteristics. J Affect Disord 2010 Sep;125(1-3):198-206. [CrossRef] [Medline]
    5. Iancu SC, Batelaan NM, Zweekhorst MBM, Bunders JFG, Veltman DJ, Penninx BWJH, et al. Trajectories of functioning after remission from anxiety disorders: 2-year course and outcome predictors. Psychol Med 2014 Feb;44(3):593-605. [CrossRef] [Medline]
    6. Yang X, Fang Y, Chen H, Zhang T, Yin X, Man J, et al. Global, regional and national burden of anxiety disorders from 1990 to 2019: results from the Global Burden of Disease Study 2019. Epidemiol Psychiatr Sci 2021 May 06;30:e36 [FREE Full text] [CrossRef] [Medline]
    7. Meier SM, Mattheisen M, Mors O, Mortensen PB, Laursen TM, Penninx BW. Increased mortality among people with anxiety disorders: total population study. Br J Psychiatry 2016 Sep;209(3):216-221 [FREE Full text] [CrossRef] [Medline]
    8. GBD 2016 DiseaseInjury IncidencePrevalence Collaborators. Global, regional, and national incidence, prevalence, and years lived with disability for 328 diseases and injuries for 195 countries, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet 2017 Sep 16;390(10100):1211-1259 [FREE Full text] [CrossRef] [Medline]
    9. Gustavsson A, Svensson M, Jacobi F, Allgulander C, Alonso J, Beghi E, CDBE2010Study Group. Cost of disorders of the brain in Europe 2010. Eur Neuropsychopharmacol 2011 Oct;21(10):718-779. [CrossRef] [Medline]
    10. Ho KP, Hunt C, Li S. Patterns of help-seeking behavior for anxiety disorders among the Chinese speaking Australian community. Soc Psychiatry Psychiatr Epidemiol 2008 Nov;43(11):872-877. [CrossRef] [Medline]
    11. Alonso J, Liu Z, Evans-Lacko S, Sadikova E, Sampson N, Chatterji S, WHO World Mental Health Survey Collaborators. Treatment gap for anxiety disorders is global: Results of the World Mental Health Surveys in 21 countries. Depress Anxiety 2018 Mar;35(3):195-208 [FREE Full text] [CrossRef] [Medline]
    12. Rüsch N, Angermeyer MC, Corrigan PW. Mental illness stigma: concepts, consequences, and initiatives to reduce stigma. Eur Psychiatry 2005 Dec;20(8):529-539. [CrossRef] [Medline]
    13. Davies MR. The stigma of anxiety disorders. Int J Clin Pract 2000;54(1):44-47. [Medline]
    14. Schofield CA, Ponzini GT. The Skidmore Anxiety Stigma Scale (SASS): A covert and brief self-report measure. J Anxiety Disord 2020 Aug;74:102259 [FREE Full text] [CrossRef] [Medline]
    15. Alonso J, Buron A, Bruffaerts R, He Y, Posada-Villa J, Lepine J, World Mental Health Consortium. Association of perceived stigma and mood and anxiety disorders: results from the World Mental Health Surveys. Acta Psychiatr Scand 2008 Oct;118(4):305-314 [FREE Full text] [CrossRef] [Medline]
    16. Anxiety disorders. National Institute of Mental Health.   URL: https://www.nimh.nih.gov/health/topics/anxiety-disorders [accessed 2022-08-10]
    17. Yousefi R, Ghorban A. The comparison of appraisals and attitudes toward illness in schizophrenia, schizoaffective and major depression disorder. J Clin Psycol 2009;1(3):1-12. [CrossRef]
    18. Imel ZE, Steyvers M, Atkins DC. Computational psychotherapy research: scaling up the evaluation of patient-provider interactions. Psychotherapy (Chic) 2015 Mar;52(1):19-30 [FREE Full text] [CrossRef] [Medline]
    19. Shen J, Rudzicz F. Detecting anxiety through Reddit. 2017 Presented at: Proceedings of the Fourth Workshop on Computational Linguistics and Clinical Psychology: From Linguistic Signal to Clinical Reality; August; Vancouver, Canada p. 58-65. [CrossRef]
    20. Wang Y, Kraut R, Levine J. To stay or leave? the relationship of emotional and informational support to commitment in online health support groups. 2012 Presented at: Proceedings of the ACM conference on Computer Supported Cooperative Work (CSCW'12); February; New York, USA p. 833-842. [CrossRef]
    21. Giles DC, Newbold J. Self- and other-diagnosis in user-led mental health online communities. Qual Health Res 2011 Mar;21(3):419-428. [CrossRef] [Medline]
    22. Oh YS, Jung H. The Relationships between Depression and Anxiety Disorder and Online Social Media for Healthcare. Am J Health Behav 2020 Jul 01;44(4):409-419. [CrossRef] [Medline]
    23. Ríssola EA, Aliannejadi M, Crestani F. Mental disorders on online social media through the lens of language and behaviour: Analysis and visualisation. Information Processing & Management 2022 May;59(3):102890. [CrossRef]
    24. Uban A, Chulvi B, Rosso P. An emotion and cognitive based analysis of mental health disorders from social media data. Future Generation Computer Systems 2021 Nov;124:480-494. [CrossRef]
    25. Lyons M, Aksayli ND, Brewer G. Mental distress and language use: Linguistic analysis of discussion forum posts. Computers in Human Behavior 2018 Oct;87:207-211. [CrossRef]
    26. De Choudhury M, Counts S, Horvitz E, Hoff A. Characterizing and predicting postpartum depression from shared Facebook data. 2014 Presented at: Proceedings of the 17th ACM conference on Computer Supported Cooperative Work & Social Computing (CSCW'14); February; New York, USA p. 626-638. [CrossRef]
    27. Guntuku SC, Yaden DB, Kern ML, Ungar LH, Eichstaedt JC. Detecting depression and mental illness on social media: an integrative review. Current Opinion in Behavioral Sciences 2017 Dec;18:43-49. [CrossRef]
    28. Liu J, Shi M. What Are the Characteristics of User Texts and Behaviors in Chinese Depression Posts? Int J Environ Res Public Health 2022 May 18;19(10):6129 [FREE Full text] [CrossRef] [Medline]
    29. Primack BA, Shensa A, Escobar-Viera CG, Barrett EL, Sidani JE, Colditz JB, et al. Use of multiple social media platforms and symptoms of depression and anxiety: A nationally-representative study among U.S. young adults. Computers in Human Behavior 2017 Apr;69:1-9. [CrossRef]
    30. Youngmann B, Yom-Tov E. Anxiety and information seeking: evidence from large-scale mouse tracking. 2018 Apr Presented at: Proceedings of the 2018 World Wide Web Conference; April; Lyon, France p. 753-762. [CrossRef]
    31. Ahmed A, Aziz S, Toro CT, Alzubaidi M, Irshaidat S, Serhan HA, et al. Machine learning models to detect anxiety and depression through social media: A scoping review. Comput Methods Programs Biomed Update 2022;2:100066 [FREE Full text] [CrossRef] [Medline]
    32. Ta N, Li K, Yang Y, Jiao F, Tang Z, Li G. Evaluating Public Anxiety for Topic-Based Communities in Social Networks. IEEE Trans Knowl Data Eng 2022 Mar 1;34(3):1191-1205. [CrossRef]
    33. Yang Y, Ta N, Li K, Jiao F, Hu B, Li Z. Influential Factors on Collective Anxiety of Online Topic-Based Communities. Front Psychol 2021;12:740065 [FREE Full text] [CrossRef] [Medline]
    34. Giunti G, Claes M, Dorronzoro Zubiete E, Rivera-Romero O, Gabarron E. Analysing Sentiment and Topics Related to Multiple Sclerosis on Twitter. Stud Health Technol Inform 2020 Jun 16;270:911-915. [CrossRef] [Medline]
    35. Bai H, Yu G. A Weibo-based approach to disaster informatics: incidents monitor in post-disaster situation via Weibo text negative sentiment analysis. Nat Hazards 2016 May 30;83(2):1177-1196. [CrossRef]
    36. Li A, Huang X, Jiao D, O'Dea B, Zhu T, Christensen H. An analysis of stigma and suicide literacy in responses to suicides broadcast on social media. Asia Pac Psychiatry 2018 Mar;10(1):e12314. [CrossRef] [Medline]
    37. Li A, Jiao D, Zhu T. Detecting depression stigma on social media: A linguistic analysis. J Affect Disord 2018 May;232:358-362. [CrossRef] [Medline]
    38. Yu L, Jiang W, Ren Z, Xu S, Zhang L, Hu X. Corrigendum to "Detecting changes in attitudes toward depression on Chinese social media: A text analysis": [Journal of Affective Disorders 280 (2021) 354-363]. J Affect Disord 2021 Feb 15;281:994-995. [CrossRef] [Medline]
    39. Li A, Jiao D, Zhu T. Stigmatizing Attitudes Across Cybersuicides and Offline Suicides: Content Analysis of Sina Weibo. J Med Internet Res 2022 Apr 08;24(4):e36489 [FREE Full text] [CrossRef] [Medline]
    40. Li A, Jiao D, Liu X, Zhu T. A Comparison of the Psycholinguistic Styles of Schizophrenia-Related Stigma and Depression-Related Stigma on Social Media: Content Analysis. J Med Internet Res 2020 Apr 21;22(4):e16470 [FREE Full text] [CrossRef] [Medline]
    41. Pennebaker J, Booth R, Boyd R, Francis M. Linguistic inquiry and word count: LIWC2015.   URL: https://www.researchgate.net/publication/337731895_Linguistic_Inquiry_and_Word_Count_LIWC2015 [accessed 2023-03-22]
    42. Krippendorff K. Content Analysis: An Introduction to Its Methodology. Thousand Oaks, CA: SAGE Publications; 2019.
    43. Reavley NJ, Pilkington PD. Use of Twitter to monitor attitudes toward depression and schizophrenia: an exploratory study. PeerJ 2014;2:e647 [FREE Full text] [CrossRef] [Medline]
    44. Lachmar EM, Wittenborn AK, Bogen KW, McCauley HL. #MyDepressionLooksLike: Examining Public Discourse About Depression on Twitter. JMIR Ment Health 2017 Oct 18;4(4):e43 [FREE Full text] [CrossRef] [Medline]
    45. Lafferty J, Blei D. Correlated topic models. 2005 Presented at: Proceedings of the 18th International Conference on Neural Information Processing Systems (NIPS'05); December; Vancouver, British Columbia, Canada p. 147-154. [CrossRef]
    46. Griffiths TL, Steyvers M, Tenenbaum JB. Topics in semantic representation. Psychol Rev 2007 Apr;114(2):211-244. [CrossRef] [Medline]
    47. Franz PJ, Nook EC, Mair P, Nock MK. Using Topic Modeling to Detect and Describe Self-Injurious and Related Content on a Large-Scale Digital Platform. Suicide Life Threat Behav 2020 Feb;50(1):5-18. [CrossRef] [Medline]
    48. Jo W, Lee J, Park J, Kim Y. Online Information Exchange and Anxiety Spread in the Early Stage of the Novel Coronavirus (COVID-19) Outbreak in South Korea: Structural Topic Model and Network Analysis. J Med Internet Res 2020 Jun 02;22(6):e19455 [FREE Full text] [CrossRef] [Medline]
    49. Blei DM. Probabilistic topic models. Commun. ACM 2012 Apr 01;55(4):77-84. [CrossRef]
    50. Paul MJ, Dredze M. Discovering health topics in social media using topic models. PLoS One 2014;9(8):e103408 [FREE Full text] [CrossRef] [Medline]
    51. Sik D, Németh R, Katona E. Topic modelling online depression forums: beyond narratives of self-objectification and self-blaming. J Ment Health 2021 Sep 28:1-10. [CrossRef] [Medline]
    52. Low DM, Rumker L, Talkar T, Torous J, Cecchi G, Ghosh SS. 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 Oct 12;22(10):e22635 [FREE Full text] [CrossRef] [Medline]
    53. Elbattah M, Arnaud É, Gignon M, Dequen G. The role of text analytics in health care: a review of recent developments and applications. 2021 Presented at: Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies; February 11-13; Virtual p. 825-832. [CrossRef]
    54. Le Glaz A, Haralambous Y, Kim-Dufor D, Lenca P, Billot R, Ryan TC, et al. Machine Learning and Natural Language Processing in Mental Health: Systematic Review. J Med Internet Res 2021 May 04;23(5):e15708 [FREE Full text] [CrossRef] [Medline]
    55. Whitehead LC. Methodological and ethical issues in Internet-mediated research in the field of health: an integrated review of the literature. Soc Sci Med 2007 Aug;65(4):782-791. [CrossRef] [Medline]
    56. TextMind: A Chinese Language Psychological Analysis System.   URL: http://ccpl.psych.ac.cn/textmind/ [accessed 2022-08-10]
    57. Gao R, Hao B, Li H, Gao Y, Zhu T. Developing simplified Chinese psychological linguistic analysis dictionary for Microblog. 2013 Presented at: International Conference on Brain & Health Informatics (BHI'13); October; Maebashi, Japan p. 359-368. [CrossRef]
    58. Yan X, Guo J, Lan Y, Cheng X. A biterm topic model for short texts. 2013 Presented at: Proceedings of the 22nd International Conference on World Wide Web; May; Rio de Janeiro, Brazil p. 1445-1456. [CrossRef]
    59. Robert H, David S. Time Series Analysis and Its Applications: With R Examples (3rd Edition). New York, NY: Springer; 2011.
    60. Schwartz H, Eichstaedt J, Kern M, Park G, Sap M, Stillwell D, et al. Towards assessing changes in degree of depression through Facebook. 2014 Jul 27 Presented at: Proceedings of the Workshop on Computational Linguistics and Clinical Psychology: From Linguistic Signal to Clinical Reality; 2014; Baltimore, Maryland, USA p. 118-125. [CrossRef]
    61. Bathina KC, Ten Thij M, Lorenzo-Luaces L, Rutter LA, Bollen J. Individuals with depression express more distorted thinking on social media. Nat Hum Behav 2021 Apr;5(4):458-466. [CrossRef] [Medline]
    62. Nguyen T. Interaction Involvement in Cross-Culture Computer-Mediated Communication: Examination of a Communication Process in Dyadic Instant Messaging Conversations. Ann Arbor, MI: Proquest LLC; 2013.
    63. Verberne S, Batenburg A, Sanders R, van Eenbergen M, Das E, Lambooij MS. Analyzing Empowerment Processes Among Cancer Patients in an Online Community: A Text Mining Approach. JMIR Cancer 2019 Apr 17;5(1):e9887 [FREE Full text] [CrossRef] [Medline]
    64. The 50th statistical report on China's internet development. China Internet Network Information Center.   URL: http://www.cnnic.net.cn/index.html [accessed 2022-10-20]

    ‎
    Affect: affective process
    Bio: biological process
    BTM: biterm topic model
    CogMech: cognitive process
    LDA: Latent Dirichlet Allocation
    LIWC: Linguistic Inquiry and Word Count
    NegEmo: negative emotion
    Percept: perceptual process
    PosEmo: positive emotion
    Social: social process

    Edited by A Mavragani; submitted 16.01.23; peer-reviewed by M Tummalacherla, S Shaikh, M Elbattah; comments to author 27.02.23; revised version received 07.03.23; accepted 08.03.23; published 04.04.23

    Copyright

    ©Jianghong Zhu, Zepeng Li, Xiu Zhang, Zhenwen Zhang, Bin Hu. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 04.04.2023.

    This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.