用蒙特卡洛树搜索提升知识库问答的准确性与效率
KBQA-o1: Agentic Knowledge Base Question Answering with Monte Carlo Tree Search
- 基于ReAct框架和树搜索实现逐步推理
- 在有限标注数据下将F1提升至78.5%
- 适合低资源场景下的智能问答系统研究
知识库问答（KBQA）旨在利用大规模结构化知识库回答自然语言问题。尽管大语言模型（LLMs）取得进展，KBQA仍面临知识库感知弱、效果与效率失衡、依赖标注数据等问题。为此，我们提出KBQA-o1，一种基于蒙特卡洛树搜索（MCTS）的智能体式KBQA方法。该方法采用基于ReAct的代理流程，通过逐步生成逻辑形式并探索知识库环境；同时引入由策略与奖励模型驱动的MCTS，平衡智能体探索的性能与搜索空间。借助启发式探索，KBQA-o1生成高质量标注数据，支持增量微调。实验表明，在有限标注数据条件下，该方法优于现有低资源KBQA方法：使用Llama-3.1-8B模型时，GrailQA的F1达到78.5%，相较之前最优方法（使用GPT-3.5-turbo）的48.5%显著提升。代码已公开。
原文摘要 · Abstract (English)
Knowledge Base Question Answering (KBQA) aims to answer natural language questions with a large-scale structured knowledge base (KB). Despite advancements with large language models (LLMs), KBQA still faces challenges in weak KB awareness, imbalance between effectiveness and efficiency, and high reliance on annotated data. To address these challenges, we propose KBQA-o1, a novel agentic KBQA method with Monte Carlo Tree Search (MCTS). It introduces a ReAct-based agent process for stepwise logical form generation with KB environment exploration. Moreover, it employs MCTS, a heuristic search method driven by policy and reward models, to balance agentic exploration's performance and search space. With heuristic exploration, KBQA-o1 generates high-quality annotations for further improvement by incremental fine-tuning. Experimental results show that KBQA-o1 outperforms previous low-resource KBQA methods with limited annotated data, boosting Llama-3.1-8B model's GrailQA F1 performance to 78.5% compared to 48.5% of the previous sota method with GPT-3.5-turbo. Our code is publicly available.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。