arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2502.13428cs.CLcs.AI
收藏

MCTS-KBQA:基于信息增益奖励的知识库问答蒙特卡洛树搜索

MCTS-KBQA: Monte Carlo Tree Search with Information Gain Rewards for Knowledge Base Question Answering

Peking University(北京大学)

Guanming Xiong, Haochen Li, Zonghong Dai, Liqiang Wen, Wen Zhao

AI总结:

该研究针对基于LLM的KBQA中MCTS应用的奖励设计与计算成本问题,提出Fast MCTS方法,以信息增益奖励替代中间状态终端展开,在四个KBQA基准上提升了准确率-成本权衡。

摘要:

本研究探索如何通过蒙特卡洛树搜索(MCTS)改进基于大语言模型(LLM)的知识库问答(KBQA)推理。将MCTS应用于基于LLM的KBQA仍具挑战性,因为奖励设计困难且基于展开的搜索计算成本高昂。现有MCTS风格方法要么依赖直接的LLM评分,要么需要大量数据训练单独的奖励模型,且通常仅在终端状态提供奖励。为解决这些局限,我们提出Fast MCTS,其用信息增益(IG)奖励替代中间状态的终端展开。IG奖励实现为经清理的交互历史上基于问题条件的困惑度比代理,通过开源指令LLM的前向传播计算,无需额外奖励模型训练。在四个KBQA基准上的实验表明,Fast MCTS始终优于线性基线,且相较于基于展开的经典MCTS,通常能提升准确率-成本权衡。代码和数据可在此https URL获取。

英文摘要:

This work investigates how to improve large language model (LLM)-based reasoning for knowledge base question answering (KBQA) via Monte Carlo Tree Search (MCTS). Applying MCTS to LLM-based KBQA remains challenging because reward design is difficult and rollout-based search is computationally expensive. Existing MCTS-style methods either rely on direct LLM scoring or require substantial data to train separate reward models, and they often provide rewards only at terminal states. To address these limitations, we propose Fast MCTS, which replaces terminal rollouts with an information gain (IG) reward for intermediate states. The IG reward is implemented as a question-conditioned PPL-ratio proxy over sanitized interaction histories, computed by forward passes of an open-source instruction LLM without additional reward-model training. Experiments on four KBQA benchmarks show that Fast MCTS consistently outperforms linear baselines and generally improves the accuracy-cost trade-off relative to rollout-based Classic MCTS. Code and data are available at https://github.com/JimXiongGM/MCTS-KBQA.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑