Deep Tabular Research via Continual Experience-Driven Execution

TL;DR

This paper introduces a novel agentic framework for deep tabular research, enabling large language models to perform complex multi-step reasoning over unstructured tables with hierarchical headers and layouts.

Contribution

It formalizes Deep Tabular Research as a decision-making process and proposes a hierarchical meta graph, an expectation-aware policy, and a structured memory for continual refinement.

Findings

Effective multi-step reasoning over unstructured tables demonstrated

Separation of strategic planning and execution improves reasoning accuracy

Framework outperforms existing methods on challenging benchmarks

Abstract

Large language models often struggle with complex long-horizon analytical tasks over unstructured tables, which typically feature hierarchical and bidirectional headers and non-canonical layouts. We formalize this challenge as Deep Tabular Research (DTR), requiring multi-step reasoning over interdependent table regions. To address DTR, we propose a novel agentic framework that treats tabular reasoning as a closed-loop decision-making process. We carefully design a coupled query and table comprehension for path decision making and operational execution. Specifically, (i) DTR first constructs a hierarchical meta graph to capture bidirectional semantics, mapping natural language queries into an operation-level search space; (ii) To navigate this space, we introduce an expectation-aware selection policy that prioritizes high-utility execution paths; (iii) Crucially, historical execution outcomes are synthesized into a siamese structured memory, i.e., parameterized updates and abstracted texts, enabling continual refinement. Extensive experiments on challenging unstructured tabular benchmarks verify the effectiveness and highlight the necessity of separating strategic planning from low-level execution for long-horizon tabular reasoning.