Scholarlink

Many previous studies have shown that Large Language Models (LLMs) are highly competent on many Natural Language Processing (NLP) tasks. However, a recent study showed the poor ability of LLMs to perform causal inference based on causal graphs and data. In this paper, we propose a method to enhance LLMs based on Structure Causal Model (SCM) tools. We constructed 10 causal inference tools based on SCM theory for solving 10 different types of causal questions. LLMs help to extract the information needed for causal inference tools in the causal questions, select the appropriate causal inference tools, and combine the inference results of the causal inference tools to generate the final causal question answers. The experimental results show that the method proposed in this paper outperforms the best existing methods.

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

Improving Causal Inference of Large Language Models with SCM Tools

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zhenyang Hua,
  • Shuyue Xing,
  • Huixing Jiang,
  • Chen Wei,
  • Xiaojie Wang

摘要

Many previous studies have shown that Large Language Models (LLMs) are highly competent on many Natural Language Processing (NLP) tasks. However, a recent study showed the poor ability of LLMs to perform causal inference based on causal graphs and data. In this paper, we propose a method to enhance LLMs based on Structure Causal Model (SCM) tools. We constructed 10 causal inference tools based on SCM theory for solving 10 different types of causal questions. LLMs help to extract the information needed for causal inference tools in the causal questions, select the appropriate causal inference tools, and combine the inference results of the causal inference tools to generate the final causal question answers. The experimental results show that the method proposed in this paper outperforms the best existing methods.