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

Digital Mental Health Interventions, e-Mental Health and Cyberpsychology (1639) Peer-to-Peer Support and Online Communities (674) Natural Language Processing (735) Medicine 2.0: Social Media, Open, Participatory, Collaborative Medicine (1706) Depression and Mood Disorders; Suicide Prevention (1432) mHealth for Wellness, Behavior Change and Prevention (2988) Behavior Change (744)

Published on in Vol 27 (2025)

Preprints (earlier versions) of this paper are available at https://preprints.jmir.org/preprint/60292, first published .
The Impact of Linguistic Signals on Cognitive Change in Support Seekers in Online Mental Health Communities: Text Analysis and Empirical Study

The Impact of Linguistic Signals on Cognitive Change in Support Seekers in Online Mental Health Communities: Text Analysis and Empirical Study

The Impact of Linguistic Signals on Cognitive Change in Support Seekers in Online Mental Health Communities: Text Analysis and Empirical Study

Authors of this article:

Min Li1 Author Orcid Image ;   Dongxiao Gu1 Author Orcid Image ;   Rui Li1 Author Orcid Image ;   Yadi Gu2 Author Orcid Image ;   Hu Liu3 Author Orcid Image ;   Kaixiang Su1 Author Orcid Image ;   Xiaoyu Wang4 Author Orcid Image ;   Gongrang Zhang1 Author Orcid Image
Article Authors Cited by Tweetations Metrics

    Original Paper

    1School of Management, Hefei University of Technology, Hefei, China

    2Center for Mental Health Education, University of Shanghai for Science and Technology, Shanghai, China

    3School of Management, Southeast University, Nanjing, China

    4The 1st Affiliated Hospital, Anhui University of Traditional Chinese Medicine, Hefei, China

    *these authors contributed equally

    Corresponding Author:

    Dongxiao Gu, PhD

    School of Management

    Hefei University of Technology

    193 Tunxi Road

    Hefei, 230009

    China

    Phone: 86 13866167367

    Email: [email protected]


    Background: In online mental health communities, the interactions among members can significantly reduce their psychological distress and enhance their mental well-being. The overall quality of support from others varies due to differences in people’s capacities to help others. This results in some support seekers’ needs being met, while others remain unresolved.

    Objective: This study aimed to examine which characteristics of the comments posted to provide support can make support seekers feel better (ie, result in cognitive change).

    Methods: We used signaling theory to model the factors affecting cognitive change and used consulting strategies from the offline, face-to-face psychological counseling process to construct 6 characteristics: intimacy, emotional polarity, the use of first-person words, the use of future-tense words, specificity, and language style. Through text mining and natural language processing (NLP) technology, we identified linguistic features in online text and conducted an empirical analysis using 12,868 online mental health support reply data items from Zhihu to verify the effectiveness of those features.

    Results: The findings showed that support comments are more likely to alter support seekers’ cognitive processes if those comments have lower intimacy (βintimacy=–1.706, P<.001), higher positive emotional polarity (βemotional_polarity=.890, P<.001), lower specificity (βspecificity=–.018, P<.001), more first-person words (βfirst-person=.120, P<.001), more future- and present-tense words (βfuture-words=.301, P<.001), and fewer function words (βlinguistic_style=–.838, P<.001). The result is consistent with psychotherapists’ psychotherapeutic strategy in offline counseling scenarios.

    Conclusions: Our research contributes to both theory and practice by proposing a model to reveal the factors that make support seekers feel better. The findings have significance for support providers. Additionally, our study offers pointers for managing and designing online communities for mental health.

    J Med Internet Res 2025;27:e60292

    doi:10.2196/60292

    Keywords

    mental health; online communities; cognitive change; signaling theory; text analysis 


    Background

    People with psychological distress are more likely to seek support from online social platforms [1] with the advantage of connecting with a large group of individuals [2-4]. Internet use and online communication have shown the potential to improve mental health by strengthening social network relationships [5-7]. Brief, frequent, asynchronous communication on social media can provide a high level of social support [8].

    The proliferation of online mental health communities, spurred by the advent of social media, has provided individuals with a platform for mutual support [9,10]. Within these communities, individuals experiencing difficulties can seek support by posting their concerns, while those who have undergone similar challenges can offer guidance through comments [11]. Analyzing support seekers’ responses to supportive comments reveals whether they experience a sense of improvement. This improvement in support seekers is characterized as a “positive cognitive change” [12-14].

    We call community members who provide support to support seekers as “support providers.” They offer customized assistance according to support seekers’ unique needs [15]. Numerous studies have demonstrated that such contact through online mental health platforms can effectively improve members’ mental health and relieve mental health issues [16,17]. However, some issues persist within these communities, such as a lack of psychological counselors, the fact that most responses come from laypeople and not professionals, the fact that some responses are of low quality, the lack of active guidance, and the possibility of social stigma, among others [18,19]. These issues may prevent support seekers from seeking support for their problems, and the social stigma and cyberbullying may reduce their willingness to seek online mental health support again. Such low-quality support has a detrimental effect on both support seekers’ awareness of their mental health problems and their community loyalty [20]. Therefore, it is valuable to examine linguistic features that make support seekers feel better, as these features can guide support providers in crafting high-quality support posts, ultimately contributing to a healthier and more sustainable online mental health community.

    A few researchers have studied the effects of topic consistency, language style adaptation, responder trustworthiness, emotion, and encouraging phrases on positive mental health outcomes in online mental health communities. Specific counseling techniques have been shown to increase the likelihood of positive health outcomes [21]. Peng et al [22] used the Linguistic Inquiry and Word Count (LIWC) lexicon to examine how support received by the Reddit mental health subcommunity affects satisfaction. Saha et al [23] explored the effects of topic consistency, language style adaptation, and respondent credibility of support postings on TalkLife on effective psychosocial outcomes. The characteristics of social support in mental health communities were examined by De Choudhury and De [24] from 4 perspectives: professional vocabulary, information, instrumentality, and emotion. To compare the characteristics of high-scoring and low-scoring supporting text comments, Chikersal et al [25] proposed emotions, pronouns, encouraging phrases, mental processes and behaviors, and length as variables of the strategies of supporting text comments.

    Nevertheless, this body of research has some flaws. For instance, some characteristics of comments have not been researched, such as intimacy and uniqueness. Many studies deem the characteristic of abstraction as a low-scoring characteristic of support, but the specificity of the text and the effect of validation on cognitive change have not been calculated in depth. A large amount of interactive data are needed to verify the influence relationship [26]. In addition, existing research on factors that contribute to feeling better is less theoretically orientated. The prediction ability of a feature selected based on theory is superior to that of a feature selected based on experience [27]. The features suggested (and chosen) by the theory are more explicative, and the psychological and managerial ramifications of those features are well explained.

    The majority of the care offered by a support provider is supplied in the form of text comments within the context of the online mental health community, using linguistic cues to elicit replies from other members. As a result, the quality of the linguistic signals more accurately reflects the degree of psychological support. Linguistic signals used to aid support seekers can be divided into emotional signals and informational signals according to Chen et al [28]. Affective signals are used to elicit or enhance the support seeker’s feelings, empathy, and sense of similarity with the support receiver, while information signals improve the observability of signals and reduce the cost of receiving and understanding the support. In this study, we used affective signals and information signals to study factors influencing cognitive change in users in online mental health communities.

    Signaling Theory

    This study applied signaling theory to model factors influencing cognitive change. Spence [29] first proposed signaling theory to describe how information is transmitted from a sender to a receiver by a variety of signals, all of which are meant to produce desired results. The issue of information asymmetry between parties in many economic and social contexts is addressed by signaling theory [30], where one party has knowledge of its quality or intent that the other party does not [31]. Given that linguistic communication is the main form of interpersonal communication and that individuals do not always have access to the same information, signaling theory is closely related to the context of the online exchange of social support and the information asymmetry between social support seekers and providers [29].

    The primary components of signaling theory include the sender, the receiver, and the signal. In the context of online mental health communities, the support provider is the signaler (a holder of knowledge that can aid in and alleviate psychological issues), who offers psychological support based on private information shared by a support seeker and has the intention of sharing this knowledge with the support seeker. The support seeker, in contrast, is the receiver, who needs support and posts personal information to obtain that support. Support providers use the features of the online platform to communicate, using language and signals to facilitate communication. They communicate with support seekers in an asymmetric situation in a way that will improve the seekers’ mood, namely through cognitive change. As different language elements can express various forms of information or emotions and eventually result in various outcomes, selecting appropriate language features when replying to others is crucial to help support seekers.

    Research Model and Hypotheses

    The model of this study was developed based on signaling theory, as described earlier. Support information is the signal that the support provider sends to the support seeker, who is the signal receiver. Given existing research on the support and data characteristics of online mental health communities, as well as existing research on actual offline, face-to-face psychological counseling [32], we identified 6 characteristics: intimacy, emotional polarity, the use of first-person words, the use of future-tense words, specificity, and linguistic style. Among these characteristics, intimacy, emotional polarity, the use of first-person words, and the use of future-tense words are affective signals, while specificity and linguistic style are information signals. Figure 1 shows the complete study model.

    ‎
    Figure 1. Research model. H: hypothesis.
    Intimacy

    Closeness in relationships relates to how emotionally invested and connected a person is with other participants [33,34]. It also indicates social interaction linkages. A person connects with others through interdependence, passion, and sharing [33,35,36]. By influencing social support, intimate relationships can improve mental and physical health and have a positive impact on well-being [37]. The expression of an intimate relationship can improve a person’s happiness and social interactions [38]. Intimacy has also been shown to be associated with building social capital and improving mental health [39]. A strong intimate relationship has a higher chance of receiving social support than one with poor ties [40]. Online mental health platforms provide support seekers the channel to read empowering text comments that help them feel close to others, cared for, and supported, thereby fostering cognitive improvements. Hence, we proposed the following hypothesis:

    Hypothesis 1 (H1): Intimacy has a positive impact on cognitive change in support seekers in online mental health communities.
    Emotional Polarity

    The study of sentimental ideas, emotions, and emotions expressed in the literature is known as sentiment analysis [41]. An increasingly popular subject in emotion analysis, particularly in the context of mental health, is electrical polarity [42]. Pérez-Rosas et al [43] categorized the emotions in support text comments into 5 categories and discovered that high-quality counselors are more likely to convey positive feelings than low-quality counselors in high-quality counseling. Successful treatment outcomes are linked to higher levels of positive emotions [44]. As a result, showing more positive feelings in a supporting text comment may improve support seekers’ feelings and alter their cognitive processes. Accordingly, we proposed the following hypothesis:

    H2: Positive emotional polarity has a positive impact on cognitive change in support seekers in online mental health communities.
    First-Person Words

    The use of personal pronouns is correlated with mental health; support seekers’ psychological conditions improve when they use the first-person singular pronouns less frequently [45]. According to cognitive behavioral therapy, support seekers undergo a cognitive change when their focus turns from themselves to others, that is, when they become more encouraging and compassionate toward others [46]. In other words, symptom reduction is correlated with a psycholinguistic shift during therapy from talking about oneself and the present to talking about other people and moments in time [47]. Sharma et al [32] explored the impact of self-focused pronouns (first person) on the counseling effect and discovered that as counseling progresses, the frequency of first-person pronoun use by psychotherapists increases. Online community members—who tend to be amateurs—often use the first or the third person to help others when they share their own comparable experiences or those of well-known individuals. This language encourages support seekers to empathize with one another and shift their viewpoint from their own difficulties to those of others by making the experiences of others feel familiar, as though someone else is experiencing challenges that are similar to their own. Therefore, we posited the following hypothesis:

    H3: First-person pronoun use has a positive impact on cognitive change in support seekers in online mental health communities.
    Future-Tense Words

    Research has shown that similar to the factor of self, described in the previous section, when counselors talk about the future and help support seekers shift their viewpoint toward the future, it increases the likelihood that the support seekers will experience an altered perspective [16]. Support seekers have a better chance of surviving a crisis if they focus on the future rather than on issues from the past. We further proposed that increased use of future-oriented terms may influence support seekers to alter their viewpoint toward the future, and thus encourage cognitive change, in the setting of online mental health communities. Thus, we proposed the following hypothesis:

    H4: Future temporal features have a positive impact on cognitive change in support seekers in online mental health communities.
    Specificity

    The degree to which a reference in a text can be clearly understood is referred to as its specificity. Supportive text comments with good ratings use fewer terms with an abstract connection [25]. The degree of abstraction and concreteness of online comments affects how valuable they are seen to be, which in turn influences how online users make decisions [48]. Construal level theory [49] states that people will use low-level interpretations when they perceive a target as being close or attainable but high-level interpretations when they perceive a target as being psychologically far. Low-level explanations are concrete, but high-level explanations are abstract [50]. Low-level language helps support seekers feel empathy, reduce the psychological and social distance between themselves and support providers, and become aware of the efforts being made on the part of support providers to assist them, all of which can lead to positive cognitive changes. Thus, we proposed the following hypothesis:

    H5: Specificity characteristics have a positive impact on cognitive change in support seekers in online mental health communities.
    Language Style

    Many nonverbal cues, such as facial expressions and body gestures, disappear in the online interactive environment, and more written language is needed to seek or provide support compared to offline, face-to-face psychological counseling. Therefore, the acceptance of psychological support may be significantly influenced by the language style of the comments offered as support [51]. The language used during counseling affects the results in terms of the patient’s mental health. Online, a member’s level of identification with the group and engagement should be reflected in how closely their linguistic style resembles that of the entire community [52]. The function words in text comments more accurately reflect the language style than the substance of the comment [53]. Despite being few in number, function words make up 60% of the vocabulary that people use [54]. Function words can act as markers of emotional states, social identities, and cognitive styles [55]. The listener’s social and cognitive skills must be further developed due to the increased usage of function words [56]. The rising use of function words by those providing support in online mental health communities creates a need for more effort by support seekers and might not be helpful for their development. Accordingly, we proposed the following hypothesis:

    H6: Language style characteristics have a positive impact on cognitive change in support seekers in online mental health communities.

    Therefore, drawing on signaling theory, in this study, we proposed a framework to examine which factors affect cognitive change in online mental health communities and measured their characteristics using natural language processing (NLP) techniques. This study has potential for supporters in online mental health communities. It offers suggestions for designing and managing these communities and the interactions of their members.

    Our main research contributions are as follows: First, we proposed a model to reveal factors affecting cognitive change in online mental health communities and provided empirical support. Our model was based on theories and strategies of existing counseling scenarios. Second, we extended signaling theory to psychological support scenarios in online mental health communities, proposed unique text features of psychological support from information signals and emotional signals, and verified the hypothesis. Finally, we highlighted the significance and relevance of 2 new factors, intimacy and specificity. To the best of our knowledge, these 2 variables have not been studied in online mental support scenarios. The results regarding the influence of intimacy and specificity on cognitive change conflict with previous findings and are specific in the context of psychological support in online mental health communities.


    Data Collection

    We collected data from a representative online mental health community Zhihu, which is a publicly available platform where posts, comments, and replies can be seen by everyone. In May 2023, the Zhihu platform’s question-and-answer (Q&A) subcommunity “mental health” had 5.31 million followers. The Q&A text on Zhihu related to mental health was collected using a crawler, which also collected posting information, comment support information, and reply text. In the Q&A text, we primarily analyzed comment data and reply text. We collected a rich data set of 276,528 posts and 565,813 reply comments on Zhihu’s Q&A “mental health” subcommunity as of May 2022. We obtained 12,868 (support text, reply text) pairings. Figure 2 shows a typical sample, and more samples can be found in Multimedia Appendix 1.

    ‎
    Figure 2. A typical sample of (support text, reply text) pairings.

    Dependent Variable

    Gu et al [57] used the TextCNN and the Word2vec algorithm for cognitive change classification and found the best model with a word vector length of 300 and a window size of 3. Therefore, we marked the 12,868 text items as either having or not having cognitive change using the trained cognitive change classifier that was combined with emotional information. The text items that indicated a person had undergone cognitive change received a score of 1, while the text items that indicated no change received a score of 0.

    The 12,868 reply comments’ sentences were categorized using a cognitive change classifier yielding 5439 pieces of data with cognitive change and 7681 pieces of data without cognitive change. Word frequency analyses of the comment text revealed the presence of words such as “you,” “me,” “myself,” and “is” in both categories of comments, suggesting that the majority of support providers provide support to support seekers by sharing their own experiences or making recommendations. More frequently, in posts by children, references to the children’s parents, friends, and family, as well as certain negative emotional phrases such as “depression” and “bad” appeared. However, words such as “come on,” “hug,” “hope,” and similar expressions appear more frequently in comments that do alter cognition [58,59].

    Independent Variables

    Intimacy

    Gilbert and Karahalios [60] compared intimacy-related terms in the LIWC dictionary to the words in text passages and counted the number of times they appeared to determine the level of intimacy in the passages. Intimacy was determined by Pei and Jurgens [61] by training NLP models through extensive text analysis to determine the intimacy of text comments. Pei and Jurgens [61] argued that text that makes people feel intimate, profound, and humanized has a high level of intimacy. The intimacy of interactive text in online health communities was predicted by Reuel et al [62] using the LIWC dictionary, the latent Dirichlet allocation (LDA) topic distribution, emotion categories, and other factors. To determine the intimacy of support stickers in our study, the Chinese LIWC dictionary’s words in the categories of family, friends, home, sexuality, swearing, work, leisure, money, body, and health were employed. Higher intimacy is indicated if the support giver is willing to discuss private matters, such as family and friends. As a result, the intimacy score is greater when the support text has more terms that indicate intimacy.

    Positive Emotions

    A Python package called SnowNLP is used to process the Chinese natural language. It supports Chinese natural language operations such as text classification, sentiment analysis, part-of-speech tagging, and Chinese word segmentation. In this work, emotion in support comments was determined using the SnowNLP package. We started by using the “pip” command to install the SnowNLP package in a Python environment. The support comment text was then used as a parameter by the SnowNLP algorithm to determine emotional polarity values. Results from SnowNLP vary from 0 to 1, with values closer to 1 denoting pleasant, positive emotions and values closer to 0 denoting negative emotions.

    First-Person Words

    The point of view of posts in an online mental health community was calculated by Lyons et al [63] using the personal pronouns in the LIWC dictionary, such as “me,” “you,” and “them.” In the context of online mental health assistance, we investigated how the language viewpoint of the support provider affects cognitive change in the support seeker. First, preprocessing of the 12,868 comments permitted text segmentation, and part-of-speech tagging was used to annotate the word segmentation results. Using personal pronouns from the dictionary, terms relating to the category of personal pronouns were screened.

    Future-Tense Words

    To discuss the impact of using a temporal perspective on cognitive change in support seekers in an online mental health community, we divided time into the past, present, and future. A dictionary was created primarily through part-of-speech tagging by extracting words designated as time related. The dictionary was then split into a past-tense word dictionary, a present-tense word dictionary, and a future-tense word dictionary through further manual screening. Finally, the dictionary statistics comments confirmed the text’s distribution of different time words. First, preprocessing of the 12,868 text items permitted text segmentation, and part-of-speech tagging was used to annotate the word segmentation results. The dictionaries created, as previously described, were then used to count the occurrence of the 3 categories of time-related words in the text items. Table 1 lists several time-related terms.

    Table 1. Excerpts from the dictionary of words related to time.
    CategoryWords/phrases
    Past-tense wordsYears ago, the day before yesterday, in the past, ten years ago, childhood, at that time, then, that year, in the early days, when I was a child, before
    Present-tense wordsThe present, the moment
    Future-tense words In the future, after, a little while, tomorrow, next year, after a long time, later, within two years, at the end of the year
    Specificity

    Polander and Shalin [64] discovered that online users of mental health services use psychological language in their narratives to express particular concepts, that they assess the content of particular psychology-related posts using the categories of leisure and work from the LIWC dictionary, and that they create a narrative dictionary based on the LIWC dictionary that includes causal words, time words, emotion words, and transition words. A 62,000-word-long English specificity dictionary was created by Muraki et al [65]. Similar to Muraki et al’s [65] approach, Xie and Bi [66] developed word-level and sentence-level specificity inference models based on large-scale Chinese specificity dictionaries. In our study, based on a Chinese specificity dictionary, we calculated the influence of the level of specificity of the comment text on cognitive change in support seekers. The specificity dictionary developed by Xie and Bi [66] is presented in tabular form in their study. The dictionary was transformed into a usable data format for this investigation. Key value matching was used to determine the specificity of each word in the comment text, and the average specificity of the words was derived as a gauge of sentence specificity.

    Language Style

    Function words in text can more accurately reflect the text’s language style than its substance [53]. The use of function words in online mental health communities was examined as a method of language style calculation [66-68]. Pronouns, prepositions, adverbs, conjunctions, auxiliary verbs, etc, are examples of function words that can express significant facets of the speaker’s motivation, requirements, and personality [69]. To conduct part-of-speech tagging, count words, and calculate function words as a measure of language style, the Baidu part-of-speech tagging tool LAC [70] was used in this study. This tool allows researchers to examine the impact of support providers’ language on changes in the cognition of support seekers in an online mental health community. Only the use of prepositions, adverbs, conjunctions, and auxiliary verbs was examined in this section because personal pronouns have already been covered in depth in the section on first-person words.

    Control Variables

    Some support providers use the second-person point of view in their comments, to give direct instruction and counsel based on the support seekers’ experiences. The likelihood of a support seeker showing improvement, however, decreases if past-tense verbs are used excessively. Thus, cognitive change may be influenced by the use of both second-person and past-tense verbs.

    Yu et al [71] defined complexity as the cognitive load of knowing a certain subject, and Nadeem and Ostendorf [72] defined linguistic complexity as the difficulty faced by language users in understanding sentences. Therapists tend to use more difficult-to-read, grammatically complicated language when counseling clients who are more worried. This language use might be connected to the need for therapists to engage in deeper communication with such clients. One study calculated the complexity of the consulting language using 12 factors, such as lexical diversity, grammatical simplicity, readability, and typicality [73]. Online support seekers in online mental health communities typically describe their experiences in detail in their postings, so support providers typically do not need to engage in more in-depth communication by using language that is more challenging to read. This situation contrasts with that of face-to-face counseling.

    The mental health difficulties posted online by support seekers may change over time due to other, unidentified variables because the connections formed in the online community are asynchronous. Hence, our study used the interval between the time of the original post and the time of the supporting comments as a control variable. Additionally, those who seek support from a larger group of people may interact more frequently and feel certain psychological comforts, which influences their own cognitive improvement. As a result, this study also used the support seeker’s number of followers, the interval between the original posting and the support seeker responding to comments, and the number of supporting comments as control variables.

    In summary, this study’s control variables were the use of the second person, the use of the past tense, topic complexity, the number of likes on support comments, posting and comment support times, and the number of comments. These variables were measured as follows: We calculate the number of second-person words and past-tense words by creating dictionaries. To determine the topic complexity, Lu et al [74] calculated the complexity of sentences and words. A lexical complexity prediction approach based on a 5-component Likert scale and manual annotation was proposed by Shardlow et al [75]; this method splits lexical complexity into 5 levels. By assessing the topic complexity of the text, Shin et al [76] used the LDA model (Genism package) to determine the complexity of the analysis text. Understanding certain words, certain sentences, and the variety of topics in the supportive comment text, when combined with psychological support situations, might influence cognitive change in support seekers. The number of words is used to gauge sentence difficulty after the preprocessing and word segmentation of the comment text. The number of subjects in the text is determined to gauge the topic complexity, and the LDA text-clustering technique is used to cluster the text.

    Based on signaling theory, we constructed a mining model of the 6 independent variables: intimacy, emotional polarity, the use of first-person words, the use of future-tense words, specificity, and linguistic style. Cognitive change was used as the dependent variable. Table 2 displays the precise procedures used to calculate each variable.

    Table 2. Description of variables.
    VariablesDescription
    Dependent variable

    		Cognitive changeSupport seekers indicate in their responses to comments whether they are getting better.
    Independent variables

    		IntimacyThe number of intimate words in a support comment.

    		Emotional polarityThe emotion degree score (in the range of 0-1, where 1 is the strongest positive emotion) in a support comment.

    		First personThe number of first-person words in a support comment.

    		Future timeThe number of future-tense words in a support comment.

    		SpecificityThe degree to which a reference in a support comment can be perceived directly; the average specificity of support stickers based on the specificity dictionary.

    		Linguistic styleThe number of function words (adverbs, prepositions, conjunctions, and auxiliary verbs) in a support comment.
    Control variables

    		Second personThe number of second-person words in a support comment.

    		Past tenseThe number of past-tense verbs in a support comment.

    		ComplexityThe number of words and the number of topics in a support comment.

    		Interval timeThe calculated time between the original post and the support comment.

    		Approval numberThe number of likes for a support comment.

    Statistical Analysis

    We used the statistical analysis program IBM SPSS 20.0 to analyze the correlation of variables. After the multicollinearity test, we used binomial logistic regression to test our hypotheses, considering that the independent and control variables were continuous and the dependent variable was dichotomous. The robustness test was carried out by using other models, as well as a sample subset. Specifically, we used the probit model to test the effect of the independent variables on the dependent variable. We then randomly selected 50% of the sample and ran binomial logistic regression and probit model tests on a subset of the sample.

    Ethical Considerations

    This study was approved by the Biomedical Ethics Committee of Hefei University of Technology (approval number HFUT20240520001H). Since the postings and comments of members of online mental health communities are related to user privacy, deidentified methods were adopted in this study: (1) After matching comment postings with reply postings, only the text of the comments and reply postings was retained, and (2) data, such as links and email addresses showing personal information contained in the text, were removed.


    Empirical Results

    Multimedia Appendix 2 displays the correlation analysis results. The outcomes of the multicollinearity test in this study were tolerance and the variance inflation factor (VIF, which is the reciprocal of tolerance). After removing word complexity, no multicollinearity issue existed among the explanatory variables, because the VIF values of all of the explanatory variables were no more than 10, and thus, all explanatory variables could be used in the regression model.

    To further understand the links between variables, we used a linear regression model to assess the elements influencing cognitive change. The 6 independent variables and the control variables were all continuous variables, but the dependent variable, cognitive change, was a dichotomous variable. Therefore, we used a binomial logistic regression model to investigate the elements. The findings demonstrated that at least 1 of the included variables was statistically significant. Table 3 shows the results of the regression analysis. Table 4 displays the findings of the hypothesis testing.

    Table 3. Results of regression analysis.
    VariableβSignificanceP valueExp(B)
    Intimacy–1.7060.213<.0010.182
    Emotional polarity.8900.058<.0012.434
    First person.1200.014<.0011.127
    Future time.3010.081<.0011.351
    Specificity–.0180.005<.0010.982
    Linguistic style–.8380.116<.0010.433
    Second person.0130.016.0421.013
    Past tense–.0680.071.0391.070
    Complexity–.0020.003.4150.998
    Approval number.1×10–50.000.0071.000
    Interval time.9×10–50.000<.0011.000
    B.5420.335N/AaN/A

    aN/A: not applicable.

    H1 concerned the impact of intimacy on cognitive change among support seekers in online mental health communities. The coefficient of prediction for cognitive change in the regression analysis, however, indicated that intimacy had a significant negative effect on cognitive change (βintimacy=–1.706, P<.001). Support texts with fewer privately similar experiences are more likely to result in a change for the better for the support seeker.

    The impact of emotional polarity on support seekers was tested in H2. This hypothesis—that a significant emotional polarity score favorably influences cognitive improvement (βemotional_polarity=.890, P<.001) and that support seekers feel comfortable and supported when support providers display more positive emotions—was validated by the regression analysis, which showed that emotional polarity has a positive regression coefficient. Straightforward helpful messages of approbation and encouragement can cause a change in cognition and pleasant emotions in support seekers.

    H3 explored the effect of using first-person terms (ie, providing a self-identity of the supporter) on cognitive change in the supporter-seeker. The hypothesis was supported by the regression results, which demonstrated that the use of the first person has a significant positive impact on cognitive change (βfirst-person=.120, P<.001). When support seekers are encouraged to improve by support providers sharing their own experiences or directly making suggestions, they are positively affected, leading to cognitive change.

    The use of words related to the future is a temporal aspect, which was examined in H4. The results showed that the use of present- and future-tense words in support postings has a significant positive impact on the occurrence of change for the better in support seekers (βfuture_words=.301, P<.001). Increased use of present- and future-tense words leads to positive cognitive change in support seekers.

    H5 concerns how the degree of specificity in support posts affects whether support seekers experience a change for the better. The regression resulted in a significant negative effect of expression specificity on cognitive change (βspecificity=–.018, P<.001). This result implies that support comment postings containing more specific words do not cause support seekers to undergo a change for the better.

    H6 examined how language style traits affect the cognitive shifts that support seekers’ experience. According to the regression analysis, the use of function words has a detrimental impact on cognitive change (βlanguage_style=–.838, P<.001). The findings support the idea that increasing the use of function words may not improve support seekers’ ability to assist others in online health forums.

    Table 4. Summary of hypothesis testing.
    HypothesisSupported?
    H1: Intimacy significantly positively affects cognitive change in support seekers in online mental health communities.No
    H2: Emotional polarity scores significantly positively influence cognitive change in support seekers in online mental health communities.Yes
    H3: First-person word use significantly positively affects cognitive change in support seekers in online mental health communities.Yes
    H4: Future temporal features significantly positively affect cognitive change in support seekers in online mental health communities.Yes
    H5: Specificity characteristics significantly positively affect cognitive change in support seekers in online mental health communities.No
    H6: Language style characteristics significantly negatively affect cognitive change in support seekers in online mental health communities.Yes

    Robustness Check

    We also used the probit model to examine the effects of independent variables on the dependent variable in order to evaluate the robustness of the model. Tables 5-7 contain the test results. The positive, negative, and significant impacts of the independent variables on the dependent variable were all supported by the binomial regression model. After selecting 6434 (50%) of the 12,868 data items at random, the sample subset was subjected to binomial logistic regression and a probit model test. The test results showed that both positive and negative effects of the independent variables on the dependent variable were significant and matched the effects of the total sample.

    Table 5. Results of probit regression analysis on the full sample (N=12,868).
    VariableβSignificanceP value
    Intimacy–.9600.179<.001
    Emotional polarity.4650.037<.001
    First person.0540.007<.001
    Future time.1560.028<.001
    Specificity–.0130.003<.001
    Linguistic style–.5340.113<.001
    Second person.0130.010.196
    Past tense.0280.042.507
    Complexity.00230.002.349
    Approval number1×10–60.000.009
    Interval time8×10–60.000.001
    B.0180.043N/Aa

    aN/A: not applicable.

    Table 6. Binomial logistic regression analysis results of 50% (n=6434) of the samples.
    VariableβSignificanceP valueExp(B)
    Intimacy–1.8950.407<.0010.150
    Emotional polarity.7590.087<.0012.137
    First person.0900.018<.0011.094
    Future time.2300.067<.0011.259
    Specificity–.0200.006<.0010.980
    Linguistic style–1.1150.258<.0010.328
    Second person.0120.023.5991.012
    Past tense.1390.099.1611.149
    Complexity.0020.004.6411.002
    Approval number1×10–60.000.1921.000
    Interval time8×10–60.000.0371.000
    B.1010.099N/Aa1.106

    aN/A: not applicable.

    Table 7. Results of probit regression analysis on the full sample (N=12,868).
    VariableβSignificanceP value
    Intimacy–1.1510.252<.001
    Emotional polarity.4380.051<.001
    First person.0480.010<.001
    Future time.1390.040<.001
    Specificity–.0110.004.003
    Linguistic style–.6870.160<.001
    Second person.0090.014.491
    Past tense.0800.059.171
    Complexity.0010.002.624
    Approval number1×10–60.000.174
    Interval time8×10–60.000.039
    B.0870.061 N/Aa

    aN/A: not applicable.


    Principal Findings

    This study provides valuable insights into the linguistic factors involved in support seekers in online mental health communities feeling better. We studied the impacts of intimacy, emotional polarity, the use of first-person words, the use of future-tense words, specificity, and language style on cognitive change in support seekers in online mental health communities. We found that support comments with lower intimacy, higher positive emotional polarity, lower specificity, more first-person words, more future- and present-tense words, and fewer function words are more likely to make support seekers feel better. Our results show that the effects of most factors are consistent with the hypotheses, except for intimacy and specificity.

    According to signaling theory [28], affective signals make support seekers feel better by increasing feelings of empathy and similarity between support seekers and support providers. Our findings show that positive emotions, the use of first-person words, and the use of future-tense words can make support seekers feel better. This result indicates that support providers can make support seekers perceive similarity and develop emotional empathy through positive emotional guidance, their own similar experiences, and directing the gaze to the future so that improvement occurs. According to information signaling [28], language style makes support seekers feel better by reducing the cost of receiving and understanding support signals. This means that support providers can reduce the use of function words.

    The findings also suggest that intimacy has a negative effect on cognitive change and that support texts with less intimacy are more likely to make support seekers feel better. However, this result is contrary to existing research[37,39], which suggests that the expression of intimacy positively contributes to mental health. This result may be explained by the fact that other people in the community, such as support seekers, provide social support by expressing similar experiences [77]. These experiences stimulate the support seekers’ recollection of their own negative experiences, such as similar family environments, similar treatment experiences, or similar attitudes [78], thus preventing positive cognitive changes from occurring. However, if they encourage support seekers to elaborate on their own experiences and encourage them by expressing positive emotions, an improvement is seen, as discussed in H2.

    The possible reasons the effect of specificity on cognitive change is negative are similar. Posts with more specificity often mean that support providers disclose more details of similar negative experiences. Specific words are more likely to make it easier for support seekers to recall their own similar negative experiences in the imaginal system, which is not conducive to making them feel better.

    On this basis, we conclude that support providers providing social support by disclosing details of similar negative experiences is not conducive to supporting support seekers to feel better, and more positive emotions are needed to guide them toward the present and future. The result is consistent with psychotherapists’ psychotherapeutic strategy in offline counseling scenarios [32].

    Theoretical and Practical Implications

    This study has 2 theoretical implications. First, we transferred signaling theory to the study of factors influencing cognitive change in online mental health communities and divided the factors into affective signals and information signals. We also verified the proposed model. The findings suggest that intimacy, emotional polarity, discussion of self, use of time-related words, specificity, and language style (use of function words) all significantly influence cognitive change in support seekers in online mental health communities. Thus, this study extended signaling theory. Second, based on previous research, but also going beyond that, this study investigated how new factors, such as intimacy and specificity, affect cognitive change. It calculated the intimacy and specificity of supporting comments by creating intimacy and specificity dictionaries, and then it confirmed the impact of these 2 factors on cognitive change using regression analysis. The outcomes of this study related to intimacy and specificity disagree with other findings; we discovered that both intimacy and specificity are associated with the psychological support provided in online mental health communities and that both are adversely correlated with cognitive change.

    The research also has practical implications. Our findings encourage tailored strategies for mental health support and intervention. Online mental health community managers can use context-specific support techniques, offer real-time ideas when posting support comments, and provide information for the design of the online interface by researching variables that influence the cognitive shifts in support seekers. In addition, community support providers can customize their replies by using the best support comment tactics for each support seeker based on their unique traits. To foster relationships between support seekers and support providers, emphasis should be placed on encouraging other members of the online mental health community to write and offer assistance. Interventions can be created to allow service providers to cater their assistance to each support seeker individually. Most significantly, support seekers who have posted an issue and who are not receiving adequate support can be helped by the referral, identification, and mentoring procedures of the online health community.

    Limitations and Future Recommendations

    Our study has several limitations. First, we included 1 online mental health community, and the findings may differ depending on the data sources used. Other contexts should be examined in future studies. Second, we did not set up support strategies in an actual online mental health community to verify whether they would improve the quality of support. Future research can strategically aid auxiliary support providers in writing supportive comments by analyzing and designing strategies and combinations of strategies so that the comments will better aid support seekers and lead to cognitive change. In addition, the characteristics of support seekers should be fully considered in functional design in order to further provide personalized support strategies. For example, the features of support comments that are more agreeable to support seekers with high participation levels versus those with low participation levels should be considered, as well as variations in the features of support comments used to treat various mental health issues.

    Conclusion

    This study investigated elements that influence cognitive change in online mental health communities. It used psychological counselors’ face-to-face consultation techniques, described in existing research, to propose hypotheses, and it specifically applied signaling theory to identify factors that influence cognitive change. We found that support comments are more likely to alter support seekers’ cognitive processes if they have lower intimacy, higher positive emotional polarity, lower specificity, more first-person words, more future- and present-tense words, and fewer function words. The result is consistent with psychotherapists’ psychotherapeutic strategy in offline counseling scenarios.

    Acknowledgments

    This research was partially supported by the National Natural Science Foundation of China (grant numbers 72131006, 72271082, and 72071063) and the Natural Science Foundation of Anhui Province (grant number 2408085J041).

    Data Availability

    The data sets generated and analyzed during this study are available from the corresponding author upon reasonable request.

    Authors' Contributions

    ML and DG contributed equally. ML conceptualized the study, drafted the methodology, performed data curation and software analysis, and prepared the original draft. DG conceptualized and investigated the study. RL reviewed and edited the draft. YD completed the investigation. HL drafted the methodology and performed data curation. KX conceptualized the study. XY and GR supervised the study.

    Conflicts of Interest

    None declared.

    Multimedia Appendix 1

    Examples of support text and reply text.

    DOCX File , 13 KB
    Multimedia Appendix 2

    Relevant analysis results.

    DOCX File , 22 KB
    1. Bizzotto N, Marciano L, de Bruijn G, Schulz PJ. The empowering role of web-based help seeking on depressive symptoms: systematic review and meta-analysis. J Med Internet Res. Feb 02, 2023;25:e36964. [FREE Full text] [CrossRef] [Medline]
    2. Stefanone M, Kwon K, Lackaff D. Exploring the relationship between perceptions of social capital and enacted support online. J Comput-Mediat Comm. Sep 13, 2012;17(4):451-466. [CrossRef]
    3. Kim J. Smartphone-mediated communication vs. face-to-face interaction: two routes to social support and problematic use of smartphone. Comput Hum Behav. Feb 2017;67:282-291. [CrossRef]
    4. Vacchiano M, Valente R. Did the screens win? An autoregressive model linking leisure, relatedness and mental health. Comput Hum Behav. Jul 2021;120:106755. [CrossRef]
    5. Chen W. Internet use, online communication, and ties in Americans’ networks. Soc Sci Comput Rev. Mar 15, 2013;31(4):404-423. [CrossRef]
    6. Kawachi I, Berkman LF. Social ties and mental health. J Urban Health. Sep 2001;78(3):458-467. [FREE Full text] [CrossRef] [Medline]
    7. Vacchiano M, Bolano D. Online and offline leisure, relatedness and psychological distress: a study of young people in Switzerland. Leisure Stud. Dec 17, 2020;40(3):338-351. [CrossRef]
    8. Lu W, Hampton KN. Beyond the power of networks: differentiating network structure from social media affordances for perceived social support. New Media Soc. Jan 04, 2016;19(6):861-879. [CrossRef]
    9. Huang K, Chengalur-Smith I, Pinsonneault A. Sharing is caring: social support provision and companionship activities in healthcare virtual support communities. MISQ. Jan 1, 2019;43(2):395-423. [CrossRef]
    10. Pereira-Sanchez V, Adiukwu F, El Hayek S, Bytyçi DG, Gonzalez-Diaz JM, Kundadak GK, et al. COVID-19 effect on mental health: patients and workforce. Lancet Psychiatry. Jun 2020;7(6):e29-e30. [FREE Full text] [CrossRef] [Medline]
    11. Park A, Conway M. Harnessing Reddit to understand the written-communication challenges experienced by individuals with mental health disorders: analysis of texts from mental health communities. J Med Internet Res. Apr 10, 2018;20(4):e121. [FREE Full text] [CrossRef] [Medline]
    12. Pruksachatkun Y, Pendse S, Sharma A. Moments of change: analyzing peer-based cognitive support in online mental health forums. 2019. Presented at: CHI '19: CHI Conference on Human Factors in Computing Systems Glasgow Scotland Uk May 4 - 9, 2019; May 4-9, 2019:1-13; Glasgow, Scotland. [CrossRef]
    13. Gu D, Wang Q, Chai Y, Yang X, Zhao W, Li M, et al. Identifying the risk factors of allergic rhinitis based on Zhihu comment data using a topic-enhanced word-embedding model: mixed method study and cluster analysis. J Med Internet Res. Feb 22, 2024;26:e48324. [FREE Full text] [CrossRef] [Medline]
    14. Gu D, Liu H, Zhao H, Yang X, Li M, Liang C. A deep learning and clustering‐based topic consistency modeling framework for matching health information supply and demand. J Assoc Inf Sci Technol. Nov 2023;75(2):152-166. [CrossRef]
    15. de Waal FBM, Preston SD. Mammalian empathy: behavioural manifestations and neural basis. Nat Rev Neurosci. Aug 2017;18(8):498-509. [CrossRef] [Medline]
    16. Althoff T, Clark K, Leskovec J. Large-scale analysis of counseling conversations: an application of natural language processing to mental health. TACL. Dec 2016;4:463-476. [CrossRef]
    17. Sharma A, Lin IW, Miner AS, Atkins DC, Althoff T. Human–AI collaboration enables more empathic conversations in text-based peer-to-peer mental health support. Nat Mach Intell. Jan 23, 2023;5(1):46-57. [CrossRef]
    18. Park A, Conway M, Chen AT. Examining thematic similarity, difference, and membership in three online mental health communities from Reddit: a text mining and visualization approach. Comput Hum Behav. Jan 2018;78:98-112. [FREE Full text] [CrossRef] [Medline]
    19. Wadden D, August T, Li Q, Althoff T. The effect of moderation on online mental health conversations. 2021. Presented at: ICWSM-2021: 15th International AAAI Conference on Web and Social Media; June 8-10, 2021:751-763; Virtual. [CrossRef]
    20. Yang X, Gu D, Li H, Liang C, Jain H, Li P. Mobile health community loyalty development process in China: an empirical study from information seeking perspective. ITP. Feb 20, 2023;37(2):635-661. [CrossRef]
    21. Perez-Rosas V, Sun X, Li C, Wang Y, Resnicow K, Mihalcea R. Analyzing the quality of counseling conversations: the tell-tale signs of high-quality counseling. 2018. Presented at: Eleventh International Conference on Language Resources and Evaluation (LREC 2018); May 7-12, 2018; Miyazaki, Japan.
    22. Peng Z, Ma X, Yang D, Tsang K, Guo Q. Effects of support-seekers? community knowledge on their expressed satisfaction with the received comments in mental health communities. 2021. Presented at: CHI '21: CHI Conference on Human Factors in Computing Systems; May 8-13, 2021; Yokohama, Japan. [CrossRef]
    23. Saha K, Torous J, Caine ED, De Choudhury M. Psychosocial effects of the COVID-19 pandemic: large-scale quasi-experimental study on social media. J Med Internet Res. Nov 24, 2020;22(11):e22600. [FREE Full text] [CrossRef] [Medline]
    24. De Choudhury M, De S. Mental health discourse on Reddit: self-disclosure, social support, and anonymity. 2014. Presented at: ICWSM 2014: 8th International AAAI Conference on Weblogs and Social Media; June 2-4, 2014:71-80; Ann Arbor, MI. [CrossRef]
    25. Chikersal P, Belgrave D, Doherty G, Enrique A, Palacios J, Richards D, et al. Understanding client support strategies to improve clinical outcomes in an online mental health intervention. 2020. Presented at: CHI '20: 2020 CHI Conference on Human Factors in Computing Systems; April 25-30, 2020; Honolulu HI. [CrossRef]
    26. Gu D, Li J, Li X, Liang C. Visualizing the knowledge structure and evolution of big data research in healthcare informatics. Int J Med Inform. Feb 2017;98:22-32. [CrossRef] [Medline]
    27. Liu X, Wang GA, Fan W, Zhang Z. Finding useful solutions in online knowledge communities: a theory-driven design and multilevel analysis. Inf Syst Res. Sep 2020;31(3):731-752. [CrossRef]
    28. Chen L, Baird A, Straub D. A linguistic signaling model of social support exchange in online health communities. Decis Support Syst. Mar 2020;130:113233. [CrossRef]
    29. Spence M. Job market signaling. Q J Econ. Aug 1973;87(3):355. [CrossRef]
    30. Feng X, Hu Y, Pfaff H, Liu S, Xie J, Zhang Z. Exploring client preferences for psychological counselors in a Chinese online health community: longitudinal study. J Med Internet Res. Oct 10, 2024;26:e58428. [FREE Full text] [CrossRef] [Medline]
    31. Chen L, Rai A, Chen W, Guo X. Signaling effects under dynamic capacity in online matching platforms: evidence from online health consultation communities. Inf Syst Res. Mar 28, 2024:isre.2021.0150. [CrossRef]
    32. Sharma A, Miner A, Atkins D, Althoff T. A computational approach to understanding empathy expressed in text-based mental health support. WARNING: this paper contains content related to suicide and self-harm. 2020. Presented at: 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP); November 16-20, 2020; Online. URL: https://webofscience.clarivate.cn/wos/alldb/full-record/WOS:000855160705035 [CrossRef]
    33. Chiu C, Hsu M, Wang ET. Understanding knowledge sharing in virtual communities: an integration of social capital and social cognitive theories. Decis Support Syst. Dec 2006;42(3):1872-1888. [CrossRef]
    34. Fan H, Lederman R. Online health communities: how do community members build the trust required to adopt information and form close relationships? Eur J Inf Syst. Nov 14, 2017;27(1):62-89. [CrossRef]
    35. Liang C, Wang X, Gu D, Li P, Chen H, Xu Z. Smart management information systems (Smis): concept, evolution, research hotspots and applications. Data Intell. Nov 1, 2023;5(4):857-884. [CrossRef]
    36. Jiang L, Geng Z, Gu D, Guo S, Huang R, Cheng H, et al. RS-SVM machine learning approach driven by case data for selecting urban drainage network restoration scheme. Data Intell. May 9, 2023;5(2):413-437. [CrossRef]
    37. Zhang Y. Understanding the sustained use of online health communities from a self‐determination perspective. J Assoc Inf Sci Technol. Jul 07, 2015;67(12):2842-2857. [CrossRef]
    38. Kersten P, Borschel E, Neyer FJ, Mund M. The social side of personality: do affiliation and intimacy motives moderate associations of personal relationships with well-being? J Pers. Aug 2023;91(4):992-1011. [CrossRef] [Medline]
    39. Lambert A. Intimacy and social capital on Facebook: beyond the psychological perspective. New Media Soc. Jul 09, 2016;18(11):2559-2575. [CrossRef]
    40. Sutcliffe A, Binder J, Dunbar R. Activity in social media and intimacy in social relationships. Comput Hum Behav. Aug 2018;85:227-235. [CrossRef]
    41. Li N, Wu DD. Using text mining and sentiment analysis for online forums hotspot detection and forecast. Decis Support Syst. Jan 2010;48(2):354-368. [CrossRef]
    42. Ortigosa A, Martín JM, Carro RM. Sentiment analysis in Facebook and its application to e-learning. Comput Hum Behav. Feb 2014;31:527-541. [CrossRef]
    43. Perez-Rosas V, Wu X, Resnicow K, Mihalcea R. What makes a good counselor? Learning to distinguish between high-quality and low-quality counseling conversations. 2019. Presented at: 57th Annual Meeting of the Association for Computational Linguistics; July 28-August 2, 2019:926-935; Florence, Italy. [CrossRef]
    44. Gogoulou E, Boman M, Ben AF, Isacsson N, Kaldo V, Sahlgren M. Predicting treatment outcome from patient texts: the case of internet-based cognitive behavioural therapy. 2021. Presented at: EACL 2021: 16th Conference of the European Chapter of the Association for Computational Linguistics; April 19-23, 2021:575-580; Kyiv, Ukraine. URL: https://webofscience.clarivate.cn/wos/alldb/full-record/WOS:000863557000046 [CrossRef]
    45. Shapira N, Lazarus G, Goldberg Y, Gilboa-Schechtman E, Tuval-Mashiach R, Juravski D, et al. Using computerized text analysis to examine associations between linguistic features and clients' distress during psychotherapy. J Couns Psychol. Jan 2021;68(1):77-87. [CrossRef] [Medline]
    46. Arntz A, Hawke LD, Bamelis L, Spinhoven P, Molendijk ML. Changes in natural language use as an indicator of psychotherapeutic change in personality disorders. Behav Res Ther. Mar 2012;50(3):191-202. [CrossRef] [Medline]
    47. Nook EC, Hull TD, Nock MK, Somerville LH. Linguistic measures of psychological distance track symptom levels and treatment outcomes in a large set of psychotherapy transcripts. Proc Natl Acad Sci U S A. Mar 29, 2022;119(13):e2114737119. [FREE Full text] [CrossRef] [Medline]
    48. Zhejiang U, Huang L, Tan C, National UOS, Ke W, Clarkson U, et al. Helpfulness of online review content: the moderating effects of temporal and social cues. J Assoc Inf Syst. Jun 2018;19(06):503-522. [CrossRef]
    49. Trope Y, Liberman N. Construal-level theory of psychological distance. Psychol Rev. Apr 2010;117(2):440-463. [FREE Full text] [CrossRef] [Medline]
    50. Ho CKY, Ke W, Liu H, Chau PYK. Separate versus joint evaluation: the roles of evaluation mode and construal level in technology adoption. MISQ. Jun 1, 2020;44(2):725-746. [CrossRef]
    51. Rains SA. Language style matching as a predictor of perceived social support in computer-mediated interaction among individuals coping with illness. Commun Res. Jan 13, 2015;43(5):694-712. [CrossRef]
    52. Ludwig S, de Ruyter K, Mahr D, Wetzels M, Brüggen E, de Ruyck T. Take their word for it: the symbolic role of linguistic style matches in user communities. MISQ. Apr 4, 2014;38(4):1201-1217. [CrossRef]
    53. Ireland ME, Pennebaker JW. Language style matching in writing: synchrony in essays, correspondence, and poetry. J Pers Soc Psychol. Sep 2010;99(3):549-571. [CrossRef] [Medline]
    54. Chung C, Pennebaker J. The psychological functions of function words. In: Fiedler K, editor. Social Communication. New York, NY. Psychology Press; 2007:343-359.
    55. Pennebaker JW, Mehl MR, Niederhoffer KG. Psychological aspects of natural language. use: our words, our selves. Annu Rev Psychol. 2003;54:547-577. [CrossRef] [Medline]
    56. Grice HP. Logic and conversation. In: Cole P, Morgan JL, editors. Syntax and Semantics, Vol. 3, Speech Acts. New York, NY. Academic Press; 1975:41-58.
    57. Gu D, Li M, Yang X, Gu Y, Zhao Y, Liang C, et al. An analysis of cognitive change in online mental health communities: a textual data analysis based on post replies of support seekers. Inf Process Manag. Mar 2023;60(2):103192. [CrossRef]
    58. Gu D, Deng S, Zheng Q, Liang C, Wu J. Impacts of case-based health knowledge system in hospital management: the mediating role of group effectiveness. Inf Manag. Dec 2019;56(8):103162. [CrossRef]
    59. Hu D, Wang M, Gao F, Xu F, Gu J. Knowledge representation and reasoning for complex time expression in clinical text. Data Intell. Jul 1, 2022;4(3):573-598. [CrossRef]
    60. Gilbert E, Karahalios K. Predicting tie strength with social media. 2009. Presented at: CHI '22: CHI Conference on Human Factors in Computing Systems; April 29-May 5, 2022:211-220; New Orleans, LA. URL: https://webofscience.clarivate.cn/wos/alldb/full-record/WOS:000265679300027 [CrossRef]
    61. Pei J, Jurgens D. Quantifying intimacy in language. 2020. Presented at: 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP); November 16-20, 2020:5307-5326; Online. URL: https://webofscience.clarivate.cn/wos/alldb/full-record/WOS:000855160705038 [CrossRef]
    62. Reuel A, Peralta S, Sedoc J, Sherman G, Ungar L. Measuring the language of self-disclosure across corpora. 2022. Presented at: ACL 2022: 60th Annual Meeting of the Association for Computational Linguistics; May 22-27, 2022:1035-1047; Dublin, Ireland. URL: https://webofscience.clarivate.cn/wos/alldb/full-record/WOS:000828767401008 [CrossRef]
    63. Lyons M, Aksayli ND, Brewer G. Mental distress and language use: linguistic analysis of discussion forum posts. Comput Hum Behav. Oct 2018;87:207-211. [CrossRef]
    64. Polander E, Shalin VL. An analysis of narrative and figurative language within online alcoholism discussion forums. Int J Ment Health Addiction. Feb 16, 2013;11(4):458-473. [CrossRef]
    65. Muraki EJ, Abdalla S, Brysbaert M, Pexman PM. Concreteness ratings for 62,000 English multiword expressions. Behav Res Methods. Aug 2023;55(5):2522-2531. [CrossRef] [Medline]
    66. Xie Z, Bi R. Construction and inference technique of large-scale Chinese concreteness lexicon. Act Sci Nat Univrsitatis Pekin. 2022;58(1):1-6. [CrossRef]
    67. Gu D, Yang X, Deng S, Liang C, Wang X, Wu J, et al. Tracking knowledge evolution in cloud health care research: knowledge map and common word analysis. J Med Internet Res. Feb 25, 2020;22(2):e15142. [FREE Full text] [CrossRef] [Medline]
    68. Yang X, Gu D, Wu J, Liang C, Ma Y, Li J. Factors influencing health anxiety: the stimulus–organism–response model perspective. INTR. Jun 24, 2021;31(6):2033-2054. [CrossRef]
    69. Ríssola EA, Aliannejadi M, Crestani F. Mental disorders on online social media through the lens of language and behaviour: analysis and visualisation. Inf Process Manag. May 2022;59(3):102890. [CrossRef]
    70. Jiao Z, Sun S, Sun K. Chinese lexical analysis with deep bi-GRU-CRF network. arXiv. Preprint posted online 2012.
    71. Yu B, He Z, Xing A, Lustria MLA. An informatics framework to assess consumer health language complexity differences: proof-of-concept study. J Med Internet Res. May 21, 2020;22(5):e16795. [FREE Full text] [CrossRef] [Medline]
    72. Nadeem F, Ostendorf M. Estimating linguistic complexity for science texts. In: Estimating linguistic complexity for science texts. 2018. Presented at: Thirteenth Workshop on Innovative Use of NLP for Building Educational Applications; June 5, 2018:45-55; New Orleans, LA. [CrossRef]
    73. Wei J, Finn K, Templeton E, Wheatley T, Vosoughi S. Linguistic complexity loss in text-based therapy. 2021. Presented at: 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies; June 6-11, 2021:4450-4459; Online. [CrossRef]
    74. Lu C, Bu Y, Wang J, Ding Y, Torvik V, Schnaars M, et al. Examining scientific writing styles from the perspective of linguistic complexity. J Assos Inf Sci Technol. Dec 23, 2018;70(5):462-475. [CrossRef]
    75. Shardlow M, Cooper M, Zampieri M. CompLex — a new corpus for lexical complexity prediction from Likert Scale data. 2020. Presented at: 1st Workshop on Tools and Resources to Empower People with REAding DIfficulties (READI); May 11-16, 2020:57-62; Marseille, France. URL: https://aclanthology.org/2020.readi-1.9
    76. Shin D, He S, Lee GM, Whinston AB, Cetintas S, Lee K. Enhancing social media analysis with visual data analytics: a deep learning approach. MISQ. Dec 1, 2020;44(4):1459-1492. [CrossRef]
    77. Green BM, Van Horn KT, Gupte K, Evans M, Hayes S, Bhowmick A. Assessment of adaptive engagement and support model for people with chronic health conditions in online health communities: combined content analysis. J Med Internet Res. Jul 07, 2020;22(7):e17338. [FREE Full text] [CrossRef] [Medline]
    78. Hargreaves S, Bath PA, Duffin S, Ellis J. Sharing and empathy in digital spaces: qualitative study of online health forums for breast cancer and motor neuron disease (amyotrophic lateral sclerosis). J Med Internet Res. Jun 14, 2018;20(6):e222. [FREE Full text] [CrossRef] [Medline]

    ‎
    LDA: latent Dirichlet allocation
    LIWC: Linguistic Inquiry and Word Count
    NLP: natural language processing
    Q&A: question-and-answer
    VIF: variance inflation factor

    Edited by T de Azevedo Cardoso; submitted 11.05.24; peer-reviewed by M Vacchiano, A AL-Asadi; comments to author 09.09.24; revised version received 04.11.24; accepted 10.12.24; published 14.01.25.

    Copyright

    ©Min Li, Dongxiao Gu, Rui Li, Yadi Gu, Hu Liu, Kaixiang Su, Xiaoyu Wang, Gongrang Zhang. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 14.01.2025.

    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 (ISSN 1438-8871), 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.