KBQA-o1: Agentic Knowledge Base Question Answering with Monte Carlo Tree Search

Haoran Luo, Haihong E, Yikai Guo, Qika Lin, Xiaobao Wu, Xinyu Mu, Wenhao Liu, Meina Song, Yifan Zhu, Anh Tuan Luu

TL;DR: We propose KBQA-o1, an agentic KBQA method with Monte Carlo Tree Search.

Abstract: 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.

Lay Summary: In this work, we present KBQA-o1, a system that helps AI better answer complex questions using structured knowledge bases. Unlike traditional methods, it explores the knowledge step-by-step and uses a search strategy similar to AlphaGo (Monte Carlo Tree Search) to guide its decisions. This approach improves accuracy even with limited training data and works across different open-source AI models.

Link To Code: https://github.com/LHRLAB/KBQA-o1

Primary Area: Applications->Language, Speech and Dialog

Keywords: Knowledge Base Question Answering, Large Language Model, LLM Agents, Monte Carlo Tree Search

Submission Number: 4314