Abstract
Table question answering (TableQA) is a fundamental task in natural languageprocessing (NLP). The strong reasoning capabilities of large language models(LLMs) have brought significant advances in this field. However, as real-worldapplications involve increasingly complex questions and larger tables,substantial noisy data is introduced, which severely degrades reasoningperformance. To address this challenge, we focus on improving two corecapabilities: Relevance Filtering, which identifies and retains informationtruly relevant to reasoning, and Table Pruning, which reduces table size whilepreserving essential content. Based on these principles, we propose EnoTab, adual denoising framework for complex questions and large-scale tables.Specifically, we first perform Evidence-based Question Denoising by decomposingthe question into minimal semantic units and filtering out those irrelevant toanswer reasoning based on consistency and usability criteria. Then, we proposeEvidence Tree-guided Table Denoising, which constructs an explicit andtransparent table pruning path to remove irrelevant data step by step. At eachpruning step, we observe the intermediate state of the table and apply apost-order node rollback mechanism to handle abnormal table states, ultimatelyproducing a highly reliable sub-table for final answer reasoning. Finally,extensive experiments show that EnoTab achieves outstanding performance onTableQA tasks with complex questions and large-scale tables, confirming itseffectiveness.