MCTS-KBQA: Monte Carlo Tree Search for Knowledge Base Question Answering
Guanming Xiong, Haochen Li, Wen Zhao · 2025
Read MCTS-KBQA: Monte Carlo Tree Search for Knowledge Base Question Answering by Guanming Xiong, Haochen Li, Wen Zhao (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-13428.txt.
Description
Guanming Xiong and co-authors speed up tree search for answering questions from knowledge bases in this 2025 paper. A language model must chain facts across a vast graph to answer well. Searching that chain step by step with Monte Carlo tree search works, but judging each step usually means costly rollouts or a trained reward model.
Their method scores each partial chain by information gain instead. A ratio of model likelihoods, with the question kept in view, stands in for how much a step teaches. It needs only forward passes of an open model, no extra training, and it grades steps mid-path rather than only at the end. On four standard benchmarks the method beat straight-line baselines every time. It also improved the accuracy per unit of compute over classic rollout search. The code and data are public.
Catalog record
| CREATOR | Guanming Xiong, Haochen Li, Wen Zhao |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-13428