Automatic anxiety recognition method based on microblog text analysis.

Area of Science:

• Computational social science
• Psychological assessment
• Natural Language Processing

Background:

• Traditional mental health assessments, like self-report questionnaires, are accurate but time-consuming and labor-intensive.
• Social media platforms generate vast amounts of user-generated text data, offering a potential source for psychological insights.

Purpose of the Study:

• To investigate the feasibility of using machine learning to automatically predict anxiety states from social media text.
• To explore the effectiveness of analyzing digital footprints for psychological characteristic prediction.

Main Methods:

• Collected data from 1,039 Sina Weibo users, including anxiety self-assessment scores and all their original text posts.
• Utilized Simplified Chinese-Linguistic Inquiry and Word Count (SC-LIWC) for feature extraction.
• Trained and evaluated machine learning models, with XGBoostRegressor showing the best performance.

Main Results:

• The XGBoostRegressor model achieved a moderate Pearson correlation coefficient (r=0.322) between predicted and self-reported anxiety scores.
• The model demonstrated high reliability, with a correlation coefficient of r=0.72.
• The study confirmed the feasibility and effectiveness of using digital footprints for psychological prediction.

Conclusions:

• Machine learning analysis of social media text provides a viable and efficient method for predicting psychological characteristics like anxiety.
• This approach can supplement traditional assessment methods, offering a scalable solution for mental health monitoring.
• Digital footprints hold significant potential for understanding and assessing mental well-being.