Scholarlink

Using youth vaping as a use case, this paper explores the feasibility of applying the latest large language models (LLMs) to automatically annotate social media posts, especially on Reddit. Youth vaping poses serious health risks, and social media offers key insights into user behavior. This study evaluates the capacity of the latest LLMs—including GPT-o1, GPT-o3-mini, Gemini 2.0 Flash, Gemma 2, LLaMA 3.3, DeepSeek R1, and Grok-2—to detect user quit intentions and cessation stages from social media posts. Model outputs were benchmarked against human annotations using various prompting strategies. Preliminary results indicate OpenAI’s GPT-o1 is the best-performing model, followed by LLAMA 3.3; the worst was Gemma 2. Although the current preliminary results demonstrate the potential of LLMs, none of them can replace human annotators yet.

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

Using Large Language Models and Prompt Strategies to Annotate Healthcare Posts on Social Media

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Lucas Aust,
  • Anthony Fu

摘要

Using youth vaping as a use case, this paper explores the feasibility of applying the latest large language models (LLMs) to automatically annotate social media posts, especially on Reddit. Youth vaping poses serious health risks, and social media offers key insights into user behavior. This study evaluates the capacity of the latest LLMs—including GPT-o1, GPT-o3-mini, Gemini 2.0 Flash, Gemma 2, LLaMA 3.3, DeepSeek R1, and Grok-2—to detect user quit intentions and cessation stages from social media posts. Model outputs were benchmarked against human annotations using various prompting strategies. Preliminary results indicate OpenAI’s GPT-o1 is the best-performing model, followed by LLAMA 3.3; the worst was Gemma 2. Although the current preliminary results demonstrate the potential of LLMs, none of them can replace human annotators yet.