Computational text analysis of emotional expressions related to non-suicidal self-injury in Chinese social media.

Computational text analysis of emotional expressions related to non-suicidal self-injury in Chinese social media - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Sci Rep . 2026 Apr 12;16:17014. doi: 10.1038/s41598-026-48077-2 Computational text analysis of emotional expressions related to non-suicidal self-injury in Chinese social media Qi Zhou 1 , Yue Tang 2 , Shuyan Zhang 3 , Lili Yang 4 , Jiekai Yang 5 , Fange Shen 1 , Lihui Yan 6 , Guo Zhen Shen 7, ✉ Author information Article notes Copyright and License information 1 Hangzhou Linping District Integrated Traditional Chinese and Western Medicine Hospital, Hangzhou, 311100 China 2 Chemical Biology Research Center, School of Pharmaceutical Sciences, Wenzhou Medical University, Wenzhou, 325035 Zhejiang China 3 The Second Clinical College, Zhejiang Chinese Medical University, Hangzhou, 310053 China 4 Nursing education department, School of Medicine, Sir Run Run Shaw hospital of Zhejiang University, Hangzhou, 320016 China 5 Department of Nursing, International School of Medicine, the Fourth Affiliated Hospital of School of Medicine, International Institutes of Medicine, Zhejiang University, Yiwu, 322000 China 6 Nursing department, Hangzhou Linping District Integrated Traditional Chinese and Western Medicine Hospital, Hangzhou, 311100 Zhejiang China 7 Geriatric Department One, Tongde Hospital of Zhejiang Province, Hangzhou, 311100 Zhejiang China ✉ Corresponding author. Received 2026 Jan 7; Accepted 2026 Apr 6; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13230886  PMID: 41968124 Abstract Non-suicidal self-injury (NSSI) is closely associated with emotion dysregulation, yet emotional experiences may be underreported in traditional research due to cultural and social constraints. This study aimed to characterize emotional expressions related to NSSI in Chinese social media discourse. Public posts from Zhihu, Weibo, and Xiaohongshu (2019–2025) were screened, yielding 4,185 NSSI-related texts. A hybrid lexicon-based sentiment analysis classified emotions into seven categories, supplemented by unsupervised content clustering. Negative emotions comprised over 70% of expressions. Sadness was most prevalent, followed by fear, anger, and disgust. Sadness reflected chronic emotional invalidation; fear centered on disclosure and social consequences; anger was internalized and self-directed; disgust related to stigma and body alienation. Positive emotions were infrequent: happiness indicated transient relief after self-injury, acceptance emerged through peer validation, and surprise was linked to dissociative or impulsive episodes. NSSI-related emotions expressed online are predominantly negative, chronic, and socially embedded. Social media analysis provides a complementary approach for understanding emotional processes in NSSI and may inform culturally sensitive interventions. Keywords: Attitude, Health, Media, Nonsuicidal self injuries, Sentiment analysis, Computational linguistics, Mixed-methods Subject terms: Health care, Psychology, Psychology Introduction Non-suicidal self-injury (NSSI) is defined as the deliberate and direct destruction of one’s own body tissue without suicidal intent, commonly manifesting as behaviors such as cutting, burning, or hitting 1 . In recent years, NSSI has emerged as a critical public health concern in China, particularly among adolescents 1 , 2 . Meta-analytic evidence indicates that the lifetime prevalence of NSSI among Chinese social media users ranges from approximately 22% to 33%, with consistently higher rates reported in socioeconomically disadvantaged and rural populations 2 , 3 . Beyond its immediate physical consequences, NSSI is clinically significant due to its strong association with subsequent suicide attempts and its frequent comorbidity with depressive disorders, anxiety disorders, and borderline personality pathology 4 , 5 . These findings underscore the urgent need to deepen understanding of the psychological mechanisms underlying NSSI in the Chinese sociocultural context. Contemporary theoretical models emphasize the central role of emotion in the initiation and maintenance of NSSI 6 . Among these, the Affect Regulation Model posits that NSSI serves as a maladaptive strategy for alleviating overwhelming negative affect, emotional numbness, or heightened physiological arousal 7 . From this perspective, emotional attitudes—such as despair, self-blame, anger, or relief—are not merely correlates of NSSI but constitute its functional drivers 6 , 8 . However, despite the theoretical emphasis on emotional processes, empirical research has largely relied on structured questionnaires and retrospective self-report measures, which may insufficiently capture the complexity and immediacy of emotional experience associated with self-injurious behavior 9 , 10 . In the Chinese context, this limitation may be further amplified by cultural norms emphasizing social harmony, emotional restraint, and the preservation of “face” (mianzi) 11 , 12 . Disclosure of psychological distress or self-harm behaviors is often perceived as threatening to personal and family honor, potentially leading individuals to minimize or conceal their true emotional states in clinical or survey settings 12 . As a result, traditional research methods may systematically underestimate the intensity, diversity, and ambivalence of emotions experienced by individuals who engage in NSSI 11 , 13 . This methodological constraint highlights the need for alternative data sources capable of capturing more spontaneous and less socially filtered emotional expression. Social media platforms have increasingly become such a space 14 , 15 . In China, Sina Weibo functions as a major venue for emotional disclosure. Features such as anonymous posting and “Tree Hole” (shudong) communities facilitate a form of online disinhibition, enabling users to articulate distress, self-injury experiences, and emotional conflicts that are rarely expressed offline 16 . At the same time, these platforms present complex risks, including the normalization or diffusion of self-injury practices within highly active topic communities 2 . Despite these concerns, social media data offer an ecologically valid and temporally sensitive window into the lived emotional experiences surrounding NSSI, complementing traditional clinical observations 14 . This study aims to systematically analyze public emotional expressions related to non-suicidal self-injury (NSSI) on Chinese social media platforms. Rather than modeling individual-level behavioral trajectories, this research provides a computational characterization of the emotional landscape of NSSI-related digital expression. By identifying recurrent emotional patterns and thematic structures in publicly shared narratives, this study seeks to complement clinical research and inform culturally sensitive prevention and intervention strategies. Method This study adopted a five-step methodological framework, including data collection, data preprocessing, relevance verification, sentiment analysis, and content analysis. The research design aimed to systematically examine public sentiments and perceptions related to non-suicidal self-injury (NSSI) as expressed on Chinese social media platforms. Data collection Publicly accessible user-generated content was collected from three major Chinese social media platforms—Zhihu, Weibo, and Xiaohongshu—chosen for their complementary discourse characteristics. Zhihu primarily hosts knowledge-oriented and experiential discussions, Weibo reflects rapid emotional reactions and public opinion diffusion, and Xiaohongshu emphasizes personal narratives and peer-sharing related to mental health and daily life 17 . Posts dated between January 1, 2019 and October 31, 2025 were retrospectively retrieved through automated scraping conducted between July and October 2025. Data preprocessing Inclusion criteria were: (1) posts explicitly or implicitly discussing non-suicidal self-injury, self-harm behaviors without suicidal intent, or closely related psychological experiences; (2) content publicly visible at the time of data retrieval; (3) text length ≥ 12 Chinese characters to ensure sufficient semantic content.Exclusion criteria included: (1) commercial advertising, automated content, or duplicated posts; (2) content unrelated to NSSI (e.g., metaphorical uses of “self-harm” without psychological meaning); (3) image-only posts or texts containing excessive unreadable symbols. Given the sensitive nature of non-suicidal self-injury (NSSI)-related content, this study adhered to strict harm-minimization principles. All data were anonymized prior to analysis, and no attempts were made to identify individual users. Reporting avoids reproducing harmful or triggering content verbatim and focuses on aggregate patterns.The research protocol was reviewed and approved by the Institutional Review Board of XXXXXX (Approval No. 2025_XXX). The Board granted approval with a waiver of informed consent due to secondary analysis of anonymized publicly accessible textual data. Relevance verification To ensure high thematic accuracy, a two-stage relevance verification strategy was implemented.First, a lexicon-based keyword expansion approach was applied. Seed terms related to non-suicidal self-injury (e.g., cutting, burning, emotion regulation, coping through pain, scar concealment, self-punishment, relief after injury) were expanded using word embeddings trained on large-scale Chinese mental health and clinical psychology corpora. Candidate terms with cosine similarity exceeding a predefined threshold were retained to form an expanded NSSI-specific vocabulary. Second, two trained researchers independently annotated each post as “relevant” or “irrelevant” to NSSI. Relevance criteria included references to: (1) intentional self-inflicted injury without suicidal intent; (2) emotional antecedents or consequences of NSSI (e.g., relief, shame, guilt); (3) help-seeking experiences, stigma, or peer communication related to self-injury; (4) public discussions of prevention, intervention, or social understanding of NSSI. Inter-rater reliability was substantial (Cohen’s Kappa = 0.83). Discrepancies were resolved through consensus discussion, ensuring high thematic consistency. Sentiment analysis To quantify emotional patterns within NSSI-related discourse, a rule-based sentiment analysis framework tailored to Chinese social media text was employed, consisting of four stages: text preprocessing, lexicon integration, emotion classification, and validation 18 , 19 . Texts were segmented using Jieba, supplemented with a custom dictionary incorporating clinical psychology and NSSI-related terms. Noise elements such as URLs, emojis, hashtags, and redundant punctuation were removed. Stopwords were filtered, and key expressions (e.g., emotional numbness, release, self-control, being misunderstood) were normalized. Sentiment classification was conducted using a hybrid emotion lexicon integrating: (1) the Dalian University of Technology Emotion Lexicon, (2) the HOWNET affective lexicon, and (3) a customized NSSI-specific emotion lexicon derived from high-frequency corpus terms. Each lexical unit was assigned a polarity score (–1 to + 1) and categorized into four primary emotional dimensions: sadness, fear/anxiety, anger, and positive affect (e.g., relief, hope, perceived understanding). For texts containing multiple emotions, a weighted rule-based algorithm accounted for emotional proximity to NSSI-related action verbs and experiential descriptors. To quantify the impact of classification uncertainty, a misclassification propagation analysis was conducted based on the empirical confusion matrix. Simulated datasets were generated by probabilistically reassigning emotion labels according to observed misclassification rates. Emotional distributions were recomputed across 1,000 iterations to assess robustnes. The customized NSSI-specific lexicon was developed through a three-step process: (1) High-frequency term extraction: Terms appearing in the top 5% frequency range within the NSSI corpus were extracted after stopword removal. (2) Expert annotation: Two clinical psychologists and one psychiatric nurse independently labeled candidate terms according to emotional valence and category. Terms were retained if agreement exceeded 80%0.3) Consensus refinement: Disagreements were resolved through structured discussion. Candidate lexical items were extracted from the top 5% of the frequency distribution following Zipfian inflection analysis, which indicated stable domain-specific lexical concentration at this threshold. Sensitivity testing using 3% and 8% cutoffs yielded > 87% overlap in retained core terms. Three domain experts independently annotated candidate terms into seven emotional categories. Raw agreement was 86.4%, and overall Fleiss’ Kappa was 0.78 (95% CI: 0.74–0.82). Category-specific Kappa values ranged from 0.69 to 0.82 (see Supplementary Appendix S1). Confusion matrices for all seven categories are presented in Supplementary Appendix S2.3. Misclassification occurred primarily between conceptually adjacent categories, including Acceptance and Happiness (12.3%) and Disgust and Anger (14.8%). Bootstrap resampling (1,000 iterations) yielded a mean Macro F1-score of 0.742 (SD = 0.018), indicating stable classification performance. Emotional distribution rankings remained consistent across resamples (variance < 2.1%). Classification outputs should therefore be interpreted as probabilistic linguistic categorizations rather than deterministic emotional diagnoses. Terms with ≥ 80% agreement were retained a priori based on established content validity standards. All retained terms, polarity scores, weighting coefficients, and agreement metrics are fully disclosed in Supplementary Table S2. A lexicon-based approach was selected due to its transparency, interpretability, and suitability for domain-specific adaptation in relatively small to medium-sized corpora. Unlike deep contextualized models, lexicon-based methods allow explicit tracing of emotion classification rules and weighting procedures, which enhances reproducibility. However, we acknowledge that rule-based approaches may have limited capacity to capture sarcasm, metaphor, ambivalence, and rapidly evolving internet slang commonly used in youth-dominated online spaces. This methodological limitation is considered in the interpretation of findings. The four primary emotional dimensions derived from established lexicons (sadness, fear/anxiety, anger, positive affect) served as initial polarity categories. During corpus-level refinement and thematic clustering interpretation, two additional distinct negative expressions (disgust) and two differentiated positive expressions (happiness as relief; acceptance as peer validation) were identified. To validate the seven-emotion structure, parallel analysis was conducted on document-level emotion scores. Observed eigenvalues exceeded simulated random eigenvalues for the first seven factors only. Model comparison demonstrated superior fit for the seven-category model relative to the four-category baseline (ΔAIC = 128). Additionally, clustering silhouette coefficients improved from 0.48 (k = 4) to 0.62 (k = 7), supporting structural distinctiveness (Supplementary Appendix S3). Under these conditions, fine-tuning deep contextual models risks overfitting and reduced interpretability. A lexicon-based framework was therefore selected for three methodological reasons: (1) Interpretability: emotional classification rules and weighting logic remain transparent and auditable; (2) Domain adaptability: integration of an NSSI-specific lexicon allows culturally grounded semantic refinement; (3) Reproducibility: rule-based scoring minimizes stochastic variability inherent in neural architectures. Nevertheless, we acknowledge that lexicon-based methods may not fully capture sarcasm, metaphor, or contextual ambiguity, particularly within youth-oriented online discourse. Content analysis via K-means clustering To identify latent thematic structures in NSSI discourse 20 , unsupervised content analysis was conducted using K-Means clustering. Texts were vectorized using TF-IDF (unigram–bigram, max_df = 0.92, min_df = 12), followed by dimensionality reduction via truncated Singular Value Decomposition (SVD). A total of 120 components, explaining over 91% of cumulative variance, were retained. K-Means clustering was then performed (k = 7, max_iter = 300, random_state = 24). The elbow method and silhouette analysis (silhouette coefficient = 0.62) supported the selected cluster number. Each cluster was interpreted using high-weight keywords and representative excerpts. Parallel analysis was conducted on document-level emotion score vectors rather than raw TF-IDF features, ensuring that the dimensionality assessment reflects the latent emotional structure rather than lexical sparsity. Compared with sentiment distribution alone, clustering analysis provided additional explanatory value by identifying co-occurring thematic structures and contextual patterns of emotional expression. This complementary approach enables interpretation beyond marginal emotion frequencies. Internal validity indices included a silhouette coefficient of 0.62 and a Davies–Bouldin index of 0.79. Bootstrap resampling yielded a mean Adjusted Rand Index of 0.79. Ward hierarchical clustering demonstrated 84% thematic overlap with K-means assignments. These results indicate stable cluster structure across alternative algorithms (Supplementary Appendix S4). Reproducible analytical pipeline To ensure full transparency and international reproducibility, the analytical workflow was formalized into a deterministic eight-stage pipeline: (1) raw data extraction, (2) cleaning and normalization, (3) tokenization, (4) stopword filtering, (5) relevance verification, (6) lexicon-based sentiment scoring, (7) dimensional validation, and (8) clustering analysis. All derived variables, transformation logic, and scoring rules are documented in the English data dictionary and preprocessing codebook (Supplementary Materials). All scripts were executed with fixed random seeds (random_state = 24) and version-specified packages to enable independent reconstruction of the analytical process. Full Methodological Verification(Supplementary Appendix S5). Results Data overview A total of 5,003 raw posts were initially retrieved from online platforms. After removing duplicates, irrelevant content, advertisements, and posts lacking substantive emotional expression, 4,185 posts related to non-suicidal self-injury (NSSI) were retained for final analysis. Posts were grouped into 2019–2021 and 2022–2025 periods. Fear-related expressions were proportionally higher in the earlier period, while acceptance modestly increased in the later period. These findings are descriptive and do not imply causal temporal shifts. Quantitative distribution Negative emotions (sadness, fear, anger, and disgust) constituted more than 70%. Sadness was the most frequently expressed emotion, accounting for approximately one-third of all posts. Fear was the second most common category, representing about one-fifth of the corpus. Anger and disgust were moderately prevalent, together comprising roughly one-quarter of emotional expressions. In contrast, emotions reflecting positive or adaptive states were less common. Acceptance accounted for approximately one-tenth of posts, while surprise appeared infrequently. Happiness was the least represented emotion, occurring in only a small proportion of posts. Emotional prevalence estimates were accompanied by bootstrap-derived 95% confidence intervals. Across all categories, interval widths remained narrow (< 3.5%), indicating stable distributional estimates despite classification uncertainty. Qualitative interpretation Sadness was the most pervasive emotion and reflected prolonged emotional invalidation across family, school, and peer contexts. Posts commonly described accumulated interpersonal loss, persistent academic or social pressure, and diminished perceived self-worth, indicating a chronic emotional state shaped by learned helplessness rather than situational distress. Fear centered on anticipated negative consequences, including being discovered or punished by authority figures and experiencing rejection following emotional disclosure. Many users also reported fear of losing emotional control, which contributed to concealment behaviors and reinforced emotional suppression. Anger was frequently internalized due to strong social norms discouraging its expression. Participants described concerns that expressing anger would damage relationships or violate moral expectations, resulting in the redirection of anger toward the self rather than others. Disgust was closely associated with trauma, chronic humiliation, and internalized stigma related to mental illness or perceived personal failure. Negative body-related experiences further intensified self-directed disgust, positioning the body as a target of contempt and control. Happiness, though infrequent, captured short-lived relief following self-injury. This emotion reflected rapid emotional downregulation, temporary clarity, and a fleeting sense of control, underscoring the paradoxical reinforcing function of NSSI rather than sustained positive affect. Surprise typically emerged in contexts of acute emotional rupture, such as dissociation or impulsive actions under extreme stress. Users often expressed shock or detachment from their behavior, suggesting impaired emotional awareness during these episodes. Acceptance was characterized by experiences of peer validation and emotional recognition, particularly within anonymous or semi-anonymous online spaces. This emotion reflected perceived safety and nonjudgmental understanding, enabling disclosure and mitigating feelings of isolation. Discussion The seven emotions in this study identified—sadness, fear, anger, disgust, happiness, surprise, and acceptance. The findings represent patterns of emotional articulation within publicly available digital narratives. They do not constitute direct measurement of psychological states, causal mechanisms, or population prevalence. Emotional distributions should therefore be interpreted as linguistic patterns shaped by platform affordances, anonymity norms, and self-presentation dynamics. Importantly, categories with moderate inter-rater agreement (e.g., Acceptance, Surprise, Disgust) should be interpreted with caution. These categories represent conceptually adjacent or context-dependent emotional states, which are inherently more difficult to classify in short-form social media text. Sensitivity analyses indicated that while absolute proportions varied slightly under simulated misclassification, the overall emotional hierarchy remained unchanged, supporting the robustness of the study’s central conclusions. Sadness emerged as a chronic affective baseline rather than a transient emotional response, characterized by persistent emotional invalidation, accumulated interpersonal loss, and learned helplessness. This finding is consistent with extensive evidence that emotion dysregulation and sustained negative affect are core vulnerabilities for NSSI 21 – 23 . Meta-analytic research demonstrates that individuals who engage in NSSI exhibit significantly greater difficulties in identifying, accepting, and regulating emotions, particularly sadness and hopelessness, compared with non-injuring peers 23 . However, unlike traditional clinical studies that conceptualize sadness as a symptom or mood state, online narratives frame sadness as an enduring condition embedded within social environments that repeatedly invalidate emotional experience 24 . This suggests that sadness in NSSI is not only intrapsychic but also socially produced and maintained. Clinically, this underscores the importance of interventions that address long-term emotional invalidation and enhance emotional awareness and validation, particularly through trauma-informed and emotion-focused approaches 25 . Fear in online NSSI discourse was predominantly anticipatory and relational, centering on fears of discovery, emotional exposure, and loss of control rather than fear of physical injury itself. This aligns with prior qualitative research showing that individuals who self-injure often avoid disclosure due to anticipated stigma, punishment, or coercive interventions from parents, educators, or healthcare providers 26 , 27 . The present findings extend this literature by illustrating how fear becomes an organizing emotional framework that sustains secrecy and reinforces NSSI as a private coping strategy 28 . In contrast to offline clinical contexts—where fear may be underreported due to power asymmetries—online platforms enable explicit articulation of social threat and surveillance anxiety 29 . These results highlight the need for non-punitive, collaborative care models and confidential pathways to support, which may reduce fear-driven avoidance of professional help. Anger appeared primarily as self-directed hostility shaped by moralized internalization of social norms that discourage outward expression of aggression. This is consistent with longitudinal research demonstrating that anger dysregulation is closely linked to NSSI, particularly during adolescence, and that suppressed anger predicts subsequent self-injury 30 , 31 . However, online narratives uniquely reveal how anger is framed as morally unacceptable or selfish, necessitating self-punishment to restore perceived moral balance 32 . Unlike clinical assessments that often measure anger intensity or impulsivity, online discourse captures the cognitive and ethical framing of anger that drives its redirection inward 33 . Clinically, these findings support interventions that normalize anger as a legitimate emotional signal and teach adaptive expression and regulation strategies, such as those emphasized in dialectical behavior therapy 33 . Disgust reflected profound body alienation and internalized stigma, often rooted in trauma, emotional neglect, or experiences of humiliation. Such narratives position the body not merely as a site of distress but as an estranged object, suggesting disruption in embodied self-coherence rather than solely diminished self-worth. Constructions extend beyond low self-esteem and align with trauma-related objectification and dissociative distancing processes, suggesting fractured self-body integration. While disgust has received comparatively less attention in NSSI research, existing studies link emotional neglect and negative self-concept to increased self-injury risk. The present findings extend this work by showing how online users articulate disgust through metaphors of contamination and objectification, framing the body as deserving of harm 34 , 35 . In offline research, such experiences are often indirectly captured through measures of self-esteem or body dissatisfaction, whereas online narratives provide direct phenomenological accounts of bodily rejection 34 . These findings underscore the need for trauma-informed, body-oriented interventions that address internalized stigma and restore a sense of bodily ownership and compassion 35 . Expressions of happiness or joy following NSSI were not indicative of genuine well-being but instead reflected instrumental relief and short-term physiological downregulation. This pattern aligns with affect regulation and negative reinforcement models of NSSI, which posit that self-injury temporarily reduces overwhelming emotional arousal. Sentiment analysis studies further demonstrate that affect fluctuates dynamically around self-injury disclosures, with transient relief often followed by renewed distress 33 , 36 . Online narratives clarify that this “happiness” is perceived as mechanical, fleeting, and emotionally hollow, a nuance that may be obscured in cross-sectional clinical assessments 36 . Clinically, it is crucial to acknowledge the relief function of NSSI without validating its effectiveness, while teaching alternative strategies that achieve similar physiological regulation without harm 8 , 36 . Surprise was closely associated with dissociative experiences and impulsive self-injury under acute stress, with users describing a loss of volitional control or awareness during the act. This finding is consistent with prior research linking dissociation, trauma exposure, and emotion dysregulation to NSSI 37 , 38 . However, online discourse uniquely captures the subjective shock and post-hoc astonishment individuals experience, highlighting fragmented self-monitoring during dissociative states. Offline clinical assessments may underestimate the prevalence of such experiences unless dissociation is severe or explicitly screened. These findings emphasize the importance of routine assessment for dissociative symptoms and the implementation of grounding and mindfulness-based interventions to reduce impulsive self-injury 39 . Finally, acceptance emerged as the primary positive communal emotion, rooted in peer validation, emotional resonance, and perceived safety within digital environments. Consistent with social support models, acceptance reduced feelings of isolation and shame 40 . Existing research on online NSSI communities has raised concerns about normalization and contagion effects; however, the present findings highlight the protective function of non-judgmental acknowledgment and shared language. Unlike offline peer support, which is often constrained by stigma and fear of disclosure, online anonymity facilitates emotional expression and validation 14 . Acceptance emerged as a relationally mediated emotional category rooted in peer validation and perceived recognition within digital communities. Unlike transient relief following self-injury, acceptance functioned as a context-specific emotion regulation mechanism that mitigated shame and social isolation. In stigmatized offline environments where disclosure may threaten familial reputation or social standing, digital anonymity enabled reciprocal validation without immediate reputational consequences. Clinically, these findings suggest that carefully moderated, clinician-informed online communities may serve as valuable adjuncts to treatment, preserving peer resonance while minimizing the risk of harm reinforcement 24 . This form of digital acceptance differs from traditional offline support structures, which may be constrained by stigma surrounding self-injury in collectivist cultural contexts. Online anonymity reduces reputational risk, thereby enabling affective validation that might otherwise remain inaccessible. This study has several limitations. First, the data were drawn from publicly available posts on Chinese social media platforms, which may overrepresent individuals willing to disclose emotional distress online and limit generalizability to the broader NSSI population. Although the present study integrates data from multiple social media platforms to capture broad patterns of NSSI-related emotional expression, platform-specific differences were not analyzed separately. Social media platforms differ in user demographics, communication norms, and interaction mechanisms, all of which may shape how emotional experiences are expressed and interpreted. The primary aim of this study was to provide a computational characterization of the aggregate emotional landscape of NSSI-related discourse rather than to compare platform-specific dynamics. Additionally, stratifying the dataset across platforms and emotion categories could result in reduced cell sizes, potentially affecting the stability of bootstrap-based estimates. Future research may benefit from platform-stratified analytical frameworks to examine how technological affordances and community cultures influence the linguistic representation of NSSI-related emotions. Second, the absence of demographic and clinical information precluded subgroup and severity analyses, restricting interpretation to aggregate emotional patterns. Third, although a validated lexicon-based sentiment approach was employed, subtle, contextual, or mixed emotional expressions may not have been fully captured. Fourth, the cross-sectional nature of the analysis prevents inference regarding temporal or causal relationships between emotional expression and NSSI behaviors. Fifth, online discourse may be shaped by platform norms and self-presentation, which should be considered when interpreting emotional meanings. Sixth, the dataset represents a self-selecting population of social media users who publicly discuss NSSI-related experiences. Individuals who do not use social media, prefer private or encrypted communication channels, reside in rural areas with limited platform access, or avoid public disclosure due to stigma are structurally excluded from this corpus. Older adults and individuals with more concealed or severe forms of self-injury may likewise be underrepresented. Demographic characteristics were unavailable, and findings cannot be generalized to adolescents or to the broader population of individuals who engage in non-suicidal self-injury. Finally, the dataset spans the period from 2019 to 2025, which includes the COVID-19 pandemic and associated social disruptions. Although the present study does not perform time-stratified analyses or attempt to infer temporal causal relationships, it is possible that pandemic-related stressors influenced emotional expression in online discourse. However, the present study was designed as a cross-sectional structural analysis of aggregate emotional patterns rather than a longitudinal or time-series investigation. A stratified temporal comparison would require additional model specification and would extend beyond the predefined scope of this study. Future research employing dedicated time-series or phase-sensitive modeling approaches may clarify whether emotional registers of NSSI disclosure shifted across pandemic stages.In addition, the aggregation of data across 2019–2025, including the COVID-19 pandemic period, may have influenced emotional expression patterns due to shifts in social stressors, online behavior, and mental health discourse. Pandemic-related isolation, increased digital engagement, and evolving platform norms may have amplified certain emotional expressions, particularly fear and distress. Therefore, the reported emotional distributions should be interpreted as temporally aggregated patterns rather than time-invariant characteristics. Conclusion This study highlights the value of social media data in examining emotional processes associated with non-suicidal self-injury. The results indicate that NSSI-related emotions are predominantly negative and chronic, shaped by interpersonal and sociocultural factors rather than isolated affective states. While online spaces present risks, they also provide opportunities for emotional disclosure and peer validation that may be limited in offline contexts. These findings support the integration of digital data into NSSI research and suggest that emotionally validating, culturally sensitive interventions are essential for effective prevention and care. Acknowledgements We thank Yuxuan Li for technical support. Author contributions YT, LLY, JKY, FES, LHY and SYZ contributed to the conceptualization, and writing – original draft.GZS and QZ contributed to the conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, software, resources, supervision, validation, visualization, writing – original draft, writing – review & editing. Funding This study was supported by a postdoctoral grant from Hangzhou Medical and Health Science and Technology (B20262480). Data availability All analysis scripts, preprocessing pipelines, and reproducible materials are publicly available via the GitHub repository (https://github.com/sean12138/Emotional-Expressions-Related-to-Non-Suicidal-Self-Injury-in-Chinese-Social-Media.git). The repository includes a complete English-language data dictionary, annotated code, environment specifications (requirements.txt and Docker configuration), and a synthetic dataset enabling full pipeline execution. Declarations Competing interests The authors declare no competing interests. Ethics approval and consent to participate The study did not involve human subjects, and the data came from publicly available platforms for academic research. Therefore, after the Institutional Review Board’s approval, the researchers only collected comments, which did not involve Human subjects privacy and did not cause human subjects harm. The Board (The Institutional Review Board of Hangzhou Linping District Integrated Traditional Chinese and Western Medicine Hospital (No. 2025-L-038)) granted approval with a waiver of informed consent due to secondary analysis of anonymized publicly accessible textual data. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References 1. Halicka, J. & Kiejna, A. Non-suicidal self-injury (NSSI) and suicidal: Criteria differentiation. Adv. Clin. Exp. Med. 27 , 257–261. 10.17219/acem/66353 (2018). [ DOI ] [ PubMed ] [ Google Scholar ] 2. Wang, Y-J. et al. Risk factors for non-suicidal self-injury (NSSI) in adolescents: A meta-analysis. ClinicalMedicine 46. 10.1016/j.eclinm.2022.101350 (2022). [ DOI ] [ PMC free article ] [ PubMed ] 3. Qu, D. et al. Non-suicidal self-injury in Chinese population: a scoping review of prevalence, method, risk factors and preventive interventions. Lancet Reg. Health – Western Pac. 37. 10.1016/j.lanwpc.2023.100794 (2023). [ DOI ] [ PMC free article ] [ PubMed ] 4. Turner, B. J. et al. Non-suicidal self-injury with and without borderline personality disorder: Differences in self-injury and diagnostic comorbidity. Psychiatry Res. 230 , 28–35. 10.1016/j.psychres.2015.07.058 (2015). [ DOI ] [ PubMed ] [ Google Scholar ] 5. Spitzen, T. L. et al. Predicting engagement in nonsuicidal self-injury (NSSI) over the course of 12 months: the roles of borderline personality disorder pathology and emotional consequences of NSSI. J. Affect. Disord. 277 , 631–639. 10.1016/j.jad.2020.08.049 (2020). [ DOI ] [ PubMed ] [ Google Scholar ] 6. Hasking, P., Whitlock, J., Voon, D. & Rose, A. A cognitive-emotional model of NSSI: using emotion regulation and cognitive processes to explain why people self-injure. Cogn. Emot. 31 , 1543–1556. 10.1080/02699931.2016.1241219 (2017). [ DOI ] [ PubMed ] [ Google Scholar ] 7. McKenzie, K. C. & Gross, J. J. Nonsuicidal Self-Injury: An Emotion Regulation Perspective. Psychopathology 47 , 207–219. 10.1159/000358097 (2014). [ DOI ] [ PubMed ] [ Google Scholar ] 8. Morrison, T. Exploring the duality of harm: Emotion regulation and the functions of non-suicidal self injury and outward-directed injury in young people in Aotearoa (The University of Waikato, 2025). 9. Robinson, K. Emotion in non-suicidal self-injury: A contradiction between global self-reports and real-time responses. thesis. Open Access Te Herenga Waka-Victoria University of Wellington 10.26686/wgtn.16960195.v1 (2021). 10. Hamza, C. A. & Willoughby, T. Nonsuicidal Self-Injury and Affect Regulation: Recent Findings From Experimental and Ecological Momentary Assessment Studies and Future Directions. J. Clin. Psychol. 71 , 561–574. 10.1002/jclp.22174 (2015). [ DOI ] [ PubMed ] [ Google Scholar ] 11. Ban, X. et al. Suicide and non-suicidal self-injury among Chinese adolescents: a longitudinal study based on the social ecological perspective. J. Affect. Disord. 384 , 151–162. 10.1016/j.jad.2025.05.019 (2025). [ DOI ] [ PubMed ] [ Google Scholar ] 12. Yeh, Y-J., Tsai, W., Chen, J-H., Kimel, S. & Culture Family Problems, Emotion Dysregulation and Nonsuicidal Self-injury: Differences among Taiwanese and Americans. Curr. Psychol. 43 , 3514–3522. 10.1007/s12144-023-04600-6 (2024). [ Google Scholar ] 13. Zhou, X. et al. The Relationship between Ambivalence over Emotional Expression and Nonsuicidal Self-Injury Among Chinese Adolescents: A Longitudinal Moderated Mediation Model. Archives Suicide Res. 28 , 90–106. 10.1080/13811118.2022.2134068 (2024). [ DOI ] [ PubMed ] [ Google Scholar ] 14. Wu, H., Zhou, X., Chen, D., Zheng, Y. & You, J. Longitudinal association between social media exposure and nonsuicidal self-injury among adolescents: investigating the directionality by within-person effects. Curr. Psychol. 43 , 9744–9754. 10.1007/s12144-023-05128-5 (2024). [ Google Scholar ] 15. Li, H. et al. Suicidal Ideation Risk and Socio-Cultural Factors in China: A Longitudinal Study on Social Media from 2010 to 2018. Int. J. Environ. Res. Public Health . 18. 10.3390/ijerph18031098 (2021). [ DOI ] [ PMC free article ] [ PubMed ] 16. Hu, J. The Secrets of Tree Hole: Analysis of Online Interpersonal Interaction Based on @Tree Hole Account in Weibo p. 1137–1143 (Atlantis, 2021). 10.2991/assehr.k.211020.319 17. Zhou, Q. et al. Perception and sentiment analysis of palliative care in Chinese social media: Qualitative studies based on machine learning. Soc. Sci. Med. 379 , 118178. 10.1016/j.socscimed.2025.118178 (2025). [ DOI ] [ PubMed ] [ Google Scholar ] 18. Cero, I., Luo, J. & Falligant, J. M. Lexicon-Based Sentiment Analysis in Behavioral Research. Perspect. Behav. Sci. 47 , 283–310. 10.1007/s40614-023-00394-x (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Xu, L. et al. A Novel Emotion Lexicon for Chinese Emotional Expression Analysis on Weibo: Using Grounded Theory and Semi-Automatic Methods. IEEE Access. 9 , 92757–92768. 10.1109/ACCESS.2020.3009292 (2021). [ Google Scholar ] 20. Zhou, Q., Lei, Y., Du, H. & Tao, Y. Public concerns and attitudes towards autism on Chinese social media based on K-means algorithm. Sci. Rep. 13 , 1–8. 10.1038/s41598-023-42396-4 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Savla, H. Y. & Sutar, D. A Qualitative Study on Understanding the Process of Nonsuicidal Self-injury. Annals Indian Psychiatry . 8 , 217. 10.4103/aip.aip_58_24 (2024). [ Google Scholar ] 22. Jenkins, A. L. & Schmitz, M. F. The Roles of Affect Dysregulation and Positive Affect in Non-Suicidal Self-Injury. Archives Suicide Res. 16 , 212–225. 10.1080/13811118.2012.695270 (2012). [ DOI ] [ PubMed ] [ Google Scholar ] 23. Wolff, J. C. et al. Emotion dysregulation and non-suicidal self-injury: A systematic review and meta-analysis. Eur. Psychiatry . 59 , 25–36. 10.1016/j.eurpsy.2019.03.004 (2019). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Lurigio, A. J., Nesi, D. & Meyers, S. M. Nonsuicidal self injury among young adults and adolescents: Historical, cultural and clinical understandings. Social Work Mental Health . 22 , 122–148. 10.1080/15332985.2023.2264434 (2024). [ Google Scholar ] 25. Vieira, A. I. et al. Nonsuicidal self-injury, difficulties in emotion regulation, negative urgency, and childhood invalidation: A study with outpatients with eating disorders. J. Clin. Psychol. 77 , 607–628. 10.1002/jclp.23038 (2021). [ DOI ] [ PubMed ] [ Google Scholar ] 26. Fox, K. R., Bettis, A. H., Burke, T. A., Hart, E. A. & Wang, S. B. Exploring Adolescent Experiences with Disclosing Self-Injurious Thoughts and Behaviors Across Settings. Res. Child. Adolesc. Psychopathol. 50 , 669–681. 10.1007/s10802-021-00878-x (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Taliaferro, L. A. et al. I have some people who actually really care: Young sexual minority women’s lived experiences of non-suicidal self-injury disclosure. Qualitative Res. Med. Healthc. 8 , 12632. 10.4081/qrmh.2024.12632 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Giordano, A. L., Prosek, E. A., Schmit, E. L. & Schmit, M. K. Examining coping and nonsuicidal self-injury among adolescents: A profile analysis. J. Couns. Dev. 101 , 214–223. 10.1002/jcad.12459 (2023). [ Google Scholar ] 29. Douglas, B. Exploring the Use of an Online Peer Support Community for Self-Harm. dclinpsychol (University of Leeds, 2021). 30. Chen, J., Xiong, Y., Yang, L., Liang, Y. & Ren, P. Bullying victimization and self-harm in adolescents: The roles of emotion regulation and bullying peer norms. Child Abuse Negl. 160 , 107199. 10.1016/j.chiabu.2024.107199 (2025). [ DOI ] [ PubMed ] [ Google Scholar ] 31. Entwistle, C., Hoemann, K., Nightingale, S. J. & Boyd, R. L. Psychosocial dynamics of suicidality and nonsuicidal self-injury: a digital linguistic perspective. npj Mental Health Res. 4 , 28. 10.1038/s44184-025-00142-w (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Kim, M-S. & Yu, J. Text mining for identifying the nature of online questions about non-suicidal self-injury. BMC Public. Health . 22 , 1041. 10.1186/s12889-022-13480-7 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 33. Lundeen, L., Popejoy, E. K., McCall, J., Christian, D. & Humphrey, L. Veiled Pain through Anonymity: Exploring Emotions about Nonsuicidal Self-Injury on Reddit. Trauma. Couns. Resil. 2 10.33470/2997-7088.1034 (2025). 34. Ying, J. et al. Longitudinal Associations Between Negative Body Image, Self-Disgust, and Nonsuicidal Self-Injury Among Chinese Adolescents: Disentangling Between- and Within-Person Effects. J. Youth Adolescence . 54 , 439–453. 10.1007/s10964-024-02070-y (2025). [ DOI ] [ PubMed ] [ Google Scholar ] 35. Yang, B., Deng, C., Xu, Y. & Hang, N. Trajectories and Bidirectional Associations between Online Sexual Objectification and Non-suicidal Self-Injury among Chinese Adolescents. J. Youth Adolescence . 10.1007/s10964-025-02296-4 (2025). [ DOI ] [ PubMed ] [ Google Scholar ] 36. Martínez-Pastor, E., Blanco-Ruiz, M. & Feijóo, S. Digital mental health and hidden support: a qualitative analysis of non-suicidal self-injury communities on TikTok. Front. Digit. Health . 7. 10.3389/fdgth.2025.1645276 (2025). [ DOI ] [ PMC free article ] [ PubMed ] 37. Zhu, Z. & Yik, M. The emotional trajectory of non-suicidal self-injury: sentiment analysis using social media data. J. Comput. Soc. Sc . 9 , 7. 10.1007/s42001-025-00438-z (2025). [ Google Scholar ] 38. Andersson, H. et al. Emotional Dysregulation and Trauma Symptoms Mediate the Relationship Between Childhood Abuse and Nonsuicidal Self-Injury in Adolescents. Front. Psychiatry . 13. 10.3389/fpsyt.2022.897081 (2022). [ DOI ] [ PMC free article ] [ PubMed ] 39. Nester, M. S., Pierorazio, N. A., Shandler, G., Brand, B. L. & Characteristics Methods, And Functions Of Non-Suicidal Self-Injury Among Highly Dissociative Individuals. J. Trauma. Dissociation . 24 , 333–347. 10.1080/15299732.2023.2181475 (2023). [ DOI ] [ PubMed ] [ Google Scholar ] 40. Nagy, L. M. & Muehlenkamp, J. J. Negative self-perceptions and severity of NSSI: Testing the benefits and barriers model. J. Clin. Psychol. 80 , 1365–1376. 10.1002/jclp.23667 (2024). [ DOI ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Data Availability Statement All analysis scripts, preprocessing pipelines, and reproducible materials are publicly available via the GitHub repository (https://github.com/sean12138/Emotional-Expressions-Related-to-Non-Suicidal-Self-Injury-in-Chinese-Social-Media.git). The repository includes a complete English-language data dictionary, annotated code, environment specifications (requirements.txt and Docker configuration), and a synthetic dataset enabling full pipeline execution. Articles from Scientific Reports are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (1.0 MB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Digital Phenotyping of Psychological Distress in Skin Cancer Patients#805545
• Social Media Markers of Adolescents' Suicide Hazard#859806
• Surfacing the Unsaid: CUE-Bench for Affective Stance in Chinese Discourse#444694
• Internet-delivered Emotion Regulation Individual Therapy or Adolescents (IERITA) With Self-injury Within Child and Adolescent Mental Health Services: Pilot Study#1101022
• Improving Language Models for Emotion Analysis: Insights from Cognitive Science#687737