TableSeek finds exact evidence in diverse tables using stepwise search
TableSeek: Structure-Preserving Agentic Evidence Seeking over Heterogeneous Table Corpora
Artificial Intelligence
Summary
Finding tables that hold the right facts to answer questions is tricky because many tables look similar but don’t have the exact needed data. The authors created TableSeek, a tool that carefully inspects parts of tables and their structure, asking questions step-by-step rather than deciding on a match in one go. This helps it locate precise facts inside complex tables with different formats. TableSeek does this without needing special training or pre-built indexes and matches the accuracy of strong existing methods while showing exactly how it found the evidence.
What this means in practice
- •For data analysts: Locate accurate facts inside complex heterogeneous tables to verify claims or answer questions more reliably.
- •For business intelligence teams: Improve automated table searching pipelines by integrating stepwise, structure-aware search to refine evidence identification without needing retraining.
Authors
Jiaming Tian, Liyao Li, Wentao Ye, Haobo Wang, Lihua Yu, Zujie Ren, Gang Chen, Junbo Zhao
Abstract
Open-domain table retrieval seeks tables that contain sufficient evidence for answering a question or verifying a claim. Yet semantic relevance is often misleading: topically similar tables may lack the required facts, while answer-bearing evidence is often confined to a few cells whose meaning depends on surrounding schema and table context. Heterogeneous schemas, value formats, and serializations further weaken one-shot matching. We present TableSeek, a structure-preserving agentic search framework for heterogeneous table corpora. Instead of ranking tables once, an LLM agent iteratively follows sparse clues, inspects schema-preserving previews, identifies schema- and value-level mismatches, and refines its investigation. TableSeek uses cells and schemas as evidence anchors while retaining complete tables as evidence units, enabling fine-grained localization without losing the context required for interpretation and answerability checking. Without relying on retriever training or a precomputed semantic index, TableSeek produces transparent evidence-seeking trajectories and achieves competitive end-to-end performance against strong retrieval-and-reranking pipelines on heterogeneous table benchmarks. These results suggest that active, structure-preserving evidence seeking is a promising paradigm for open-domain table retrieval.