Scholarlink

This study evaluates the discourse surrounding mental health on Reddit through the lens of AI-based emotion analysis. Drawing from a comprehensive dataset comprising 7,899 posts and 54,754 comments on 10 mental health related conditions, this study combines NLP techniques such as BERTopic and lexical analysis, and the usage of GPT-4o-mini to classify emotions. The results point to clear emotional differences between posts and comments: while titles often express urgency (e.g., ‘help’), comments show empathy (e.g., ‘love’). GPT-4o-mini matched human emotion annotations with an accuracy of 52.6%, performing better with positive emotions (58%). Thematic analysis revealed that clinical diagnoses were dominant in the posts, but the comments contained supportive dialogue. This work illustrated that LLMs have potential as an avenue for monitoring mental health and well-being, while also indicating potential drawbacks with respect to ambiguous emotion classification. The study also offers practical information for the development of AI tools in the mental health space and outlines directions for future work.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fine-Grained Emotion Classification in Reddit Mental Health Posts Using LLMs

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Valery Pamela Orjuela-Llanos,
  • Ixent Galpin

摘要

This study evaluates the discourse surrounding mental health on Reddit through the lens of AI-based emotion analysis. Drawing from a comprehensive dataset comprising 7,899 posts and 54,754 comments on 10 mental health related conditions, this study combines NLP techniques such as BERTopic and lexical analysis, and the usage of GPT-4o-mini to classify emotions. The results point to clear emotional differences between posts and comments: while titles often express urgency (e.g., ‘help’), comments show empathy (e.g., ‘love’). GPT-4o-mini matched human emotion annotations with an accuracy of 52.6%, performing better with positive emotions (58%). Thematic analysis revealed that clinical diagnoses were dominant in the posts, but the comments contained supportive dialogue. This work illustrated that LLMs have potential as an avenue for monitoring mental health and well-being, while also indicating potential drawbacks with respect to ambiguous emotion classification. The study also offers practical information for the development of AI tools in the mental health space and outlines directions for future work.