Answers are grounded in this paper and cite their sources.
This study was intended to rapidly and accurately detect non-suicidal self-injury(NSSI) risk groups on social networking services (SNS) using machine learning and text mining methods. Data were collected from the textbased SNS platform, X, using the keyword “self-injury” over a one-year period. Following preprocessing and cross-validation by three researchers, 18,758 posts were labeled (1: NSSI risk posts, 0: non-risk posts). Users were placed in the NSSI risk group if they had at least one NSSI risk post. A transformer-embedding method for natural language processing was then used, followed by supervised machine learning algorithms to validate the detection of NSSI risk and non-risk groups.After the collected SNS posts were labeled, 6,613 posts (35.25%) were identified as NSSI risk posts, while 12,147 posts (64.76%) were identified as non-risk posts. Users who posted risk-related content even once were placed in the at-risk group. Consequently, among the 10,524 participants included in the study, 1,733 users (16.47%) were identified as belonging to the NSSI risk group, and 8,791 users (83.53%) as belonging to the non-risk group. Furthermore, the development of a supervised machine learning model to detect NSSI risk groups on SNS demonstrated excellent performance, achieving an accuracy of approximately 85%. This result highlights the high precision of the model in distinguishing between NSSI risk and non-risk groups on SNS.These results indicate the NSSI risk groups on SNS can be efficiently and accurately identified using machine learning, thereby providing a cost-effective and timely approach for early detection. This study advances previous research by expanding the analytical focus from individual posts to user-level accounts. This approach provides a more comprehensive understanding of self-harm risk groups, which may be challenging to identify at the postlevel, and enhances the feasibility of connecting insights to practical interventions.
No takes yet. Share an insight, caveat, or question.
Jang et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: