Predicting negative attitudes towards suicide in social media texts: prediction model development and validation...

Area of Science:

• Computational social science
• Natural language processing
• Machine learning applications in public health

Background:

• Previous studies inadequately addressed diverse negative attitudes and lacked external validation for predictive models.
• Developing accurate machine learning models for suicide-related negative attitude prediction is crucial for improving health outcomes.
• Large-scale social media analysis offers a promising avenue for understanding and predicting these attitudes.

Purpose of the Study:

• To develop a comprehensive machine learning model for predicting negative attitudes towards suicide.
• To cover a wide spectrum of negative attitudes by analyzing a large social media dataset.
• To rigorously validate the model's performance on both population and individual levels.

Main Methods:

• Analysis of 938,866 Sina Weibo posts from 2009-2020.
• Human and computer-based content analysis for feature extraction and model development on 10,000 posts.
• External validation on remaining posts (population level) and a separate dataset (individual level).

Main Results:

• Model development achieved high performance with F1=0.93 and AUC=0.97.
• Population-level validation showed statistically significant but small effect sizes for negative attitude prediction.
• Individual-level validation yielded F1=0.76 and AUC=0.74.

Conclusions:

• Machine learning prediction of negative attitudes is efficient and necessary for public health.
• External validation of predictive models is essential before real-world implementation.
• This study highlights the potential of social media data in mental health research.