Linguistic Analysis for Identifying Depression and Subsequent Suicidal Ideation on Weibo: Machine Learning Approaches

Area of Science:

• Computational linguistics
• Social media analytics
• Mental health informatics

Background:

• Depression is a prevalent mental illness often underdiagnosed.
• Early identification of suicidal ideation (SI), a key depression symptom, is crucial for intervention.
• Social media platforms offer a potential avenue for monitoring mental health trends.

Purpose of the Study:

• To investigate linguistic characteristics associated with depression and SI using machine learning.
• To assess the efficacy of analyzing social media posts for detecting depression and SI.
• To explore the potential of online community data for mental health surveillance.

Main Methods:

• Analysis of large-scale Weibo data from depression communities and control groups.
• Application of machine learning models (logistic and linear regression) utilizing simplified Chinese version of Linguistic Inquiry and Word Count (SCLIWC) features.
• Topic modeling to complement machine learning findings.

Main Results:

• SCLIWC features significantly predicted depression (Nagelkerke's R² = 0.64, F-measure = 0.78, AUC = 0.82).
• Significant differences in SI were observed between depression and control groups (t = 24.71, p < 0.001).
• Linguistic features also significantly predicted SI (Adjusted R² = 0.42, r = 0.65 for test set).

Conclusions:

• Analyzing linguistic patterns in online depression communities is an effective method for identifying depression and SI.
• Social media data provides valuable insights for mental health monitoring and early intervention.
• This approach can support public health initiatives for depression and suicide prevention.