Scholarlink

Complex Knowledge Based Question Answering (C-KBQA) is inherently more difficult than traditional KBQA tasks, as it typically cannot be resolved solely through direct queries, necessitating relational reasoning to derive answers. To address this challenge, this paper focuses on generating Large Language Model (LLM)-friendly functional queries from natural language questions based on LLMs. In addition, to effectively generate promising functional queries, entities and relations alignment methods as well as multi-LLM fusion techniques are utilized to extract accurate answers from complex textual questions. In the CCKS2024 CGQA competition, the method proposed in this paper achieved the second place with an F1 score of 0.8186, clearly demonstrating the high effectiveness and efficiency of the method.

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

LLM-Based Functional Query Generation with Multi-relation Alignment for Complex Knowledge Based Question Answering

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • He Dong,
  • Wang Song,
  • Li Hao,
  • Liu Jianzhu,
  • He Ji,
  • Li Peiyao,
  • Tao Jiang,
  • Xu Bingyu

摘要

Complex Knowledge Based Question Answering (C-KBQA) is inherently more difficult than traditional KBQA tasks, as it typically cannot be resolved solely through direct queries, necessitating relational reasoning to derive answers. To address this challenge, this paper focuses on generating Large Language Model (LLM)-friendly functional queries from natural language questions based on LLMs. In addition, to effectively generate promising functional queries, entities and relations alignment methods as well as multi-LLM fusion techniques are utilized to extract accurate answers from complex textual questions. In the CCKS2024 CGQA competition, the method proposed in this paper achieved the second place with an F1 score of 0.8186, clearly demonstrating the high effectiveness and efficiency of the method.