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

2025-01-31

AI Summary

To address key challenges in knowledge base question answering (KBQA)—including weak knowledge awareness, inefficiency-effectiveness trade-offs, and heavy reliance on strong supervision—this paper proposes the first end-to-end reasoning framework that integrates Monte Carlo Tree Search (MCTS) into a ReAct agent. The method enables deep environment interaction via policy-driven, stepwise logical form generation and knowledge-base-guided heuristic exploration; it further supports high-quality self-labeling and incremental fine-tuning under low-resource conditions. Evaluated on GrailQA, our approach achieves 78.5% F1 with Llama-3.1-8B using only limited annotated data—substantially outperforming the prior state of the art (GPT-3.5-turbo, 48.5% F1). This demonstrates MCTS’s effectiveness and novelty in enhancing both reasoning quality and generalization capability in KBQA.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

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.

Problem

• Knowledge Base Question Answering
• Data Dependency
• Efficiency and Effectiveness Balance

Innovation

• KBQA-o1
• Monte Carlo Tree Search (MCTS)
• Self-generated Training Data