![[2602.14089] TabTracer: Monte Carlo Tree Search for Complex Table Reasoning with Large Language Models](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.14089] TabTracer: Monte Carlo Tree Search for Complex Table Reasoning with Large Language Models
Summary
TabTracer introduces a novel Monte Carlo Tree Search framework for enhancing table reasoning in large language models, improving accuracy and reducing token costs significantly.
Why It Matters
As large language models become integral in natural language processing, enhancing their reasoning capabilities with structured approaches like TabTracer is crucial. This framework addresses common limitations in existing methods, such as verification and redundancy, making it a significant advancement in AI applications involving complex data interpretation.
Key Takeaways
- TabTracer improves table reasoning accuracy by up to 6.7% compared to existing models.
- The framework employs Monte Carlo Tree Search for effective state tracking and verification.
- Redundancy in processing is minimized, leading to a 59-84% reduction in token consumption.
- Step-level verification enhances reliability and reduces hallucinations in outputs.
- The approach is validated against multiple datasets, demonstrating its robustness.
Computer Science > Databases arXiv:2602.14089 (cs) [Submitted on 15 Feb 2026] Title:TabTracer: Monte Carlo Tree Search for Complex Table Reasoning with Large Language Models Authors:Zhizhao Luo, Zhaojing Luo, Meihui Zhang, Rui Mao View a PDF of the paper titled TabTracer: Monte Carlo Tree Search for Complex Table Reasoning with Large Language Models, by Zhizhao Luo and 2 other authors View PDF HTML (experimental) Abstract:Large language models (LLMs) have emerged as powerful tools for natural language table reasoning, where there are two main categories of methods. Prompt-based approaches rely on language-only inference or one-pass program generation without step-level verification. Agent-based approaches use tools in a closed loop, but verification is often local and backtracking is limited, allowing errors to propagate and increasing cost. Moreover, they rely on chain- or beam-style trajectories that are typically combinatorially redundant, leading to high token costs. In this paper, we propose TabTracer, an agentic framework that coordinates multi-step tool calls over intermediate table states, with explicit state tracking for verification and rollback. First, it enforces step-level verification with typed operations and lightweight numeric and format checks to provide reliable rewards and suppress hallucinations. Second, execution-feedback Monte Carlo Tree Search maintains a search tree of candidate table states and uses backpropagated reflection scores to guide UCB1 s...
