Predictors of Professional Responses in Nonprofit Mental Health Forums: Interpretable Machine Learning Analysis

Area of Science:

• Digital Health
• Online Communities
• Mental Health Support

Background:

• Online mental health communities enhance access to psychological support.
• Professional contributions are vital for the sustainability of these platforms.
• Limited research exists on how post characteristics influence professional responses in nonprofit mental health forums.

Purpose of the Study:

• To investigate how textual and contextual features of user inquiries affect the quantity and length of professional responses in nonprofit mental health forums.
• To provide insights for improving interactions within online mental health communities.

Main Methods:

• Analysis of 18,572 question-and-answer records from a Chinese online mental health platform.
• Utilized BERTopic for topic extraction and a BERT-based model for sentiment analysis.
• Employed Light Gradient Boosting Machine and SHAP analysis to identify key predictive features for response quantity and length.

Main Results:

• Depression topics negatively predicted response quantity, while interpersonal topics positively correlated.
• Page views and title length were key predictors of response quantity.
• Content length, sentiment, and title length significantly influenced response length. Negative emotions and longer posts correlated with increased response quantity and length.

Conclusions:

• Textual and contextual post features significantly impact professional responses in online mental health communities.
• Findings offer practical guidance for optimizing platform design and user posting strategies.
• Future research should explore response quality beyond quantity and length.