Related Experiment Video
Updated: Mar 30, 2026
08:53
Integrating Computerized Linguistic and Social Network Analyses to Capture Addiction Recovery Capital in an Online Community
Published on: May 31, 2019
5.9K
Identifying Chinese Microblog Users With High Suicide Probability Using Internet-Based Profile and Linguistic
Li Guan1, Bibo Hao2, Qijin Cheng3
1Key Lab of Behavioral Science of Chinese Academy of Sciences Institute of Psychology Chinese Academy of Sciences Beijing China ; University of Chinese Academy of Sciences Beijing China.
JMIR Mental Health
|November 7, 2015
Summary
Identifying high suicide probability users in China is feasible using online social media data. Machine learning models can screen at-risk individuals efficiently, complementing expert analysis for large-scale surveillance.
Area of Science:
- Computational psychiatry
- Social media analytics
- Public health surveillance
Background:
- Traditional suicide risk assessment is time-intensive and faces challenges in participant engagement.
- Online social media offers an efficient, less intrusive method for identifying individuals at high risk of suicide.
- Research on utilizing social media for suicide risk detection, particularly in China, is limited.
Purpose of the Study:
- To assess the feasibility and effectiveness of using Simple Logistic Regression (SLR) and Random Forest (RF) models.
- To identify high suicide probability users on Chinese microblogs.
- To analyze profile and linguistic features from online data for suicide risk prediction.
Main Methods:
- A survey of 909 Chinese microblog users was conducted using the Suicide Probability Scale (SPS).
- High-risk individuals were defined as scoring at least one SD above the mean on the SPS or its subscales.
- SLR and RF models were trained and validated using 5-fold cross-validation on profile and linguistic features.
Main Results:
- Both SLR and RF models demonstrated comparable classification performance.
- The models successfully identified over 70% of high-risk individuals for overall suicide probability and its dimensions.
- Screening Efficiency ranged from 1/4 to 1/2, with model Precision generally below 30%.
Conclusions:
- Profile and text data from Chinese microblogs can effectively identify individuals with high suicide probability.
- While model performance requires future improvement, this approach offers a promising tool for preliminary screening.
- Machine learning-based identification can enhance the efficiency of large-scale suicide probability surveillance when used alongside expert evaluation.
