The paper "CABINET: Content Relevance Based Noise Reduction for Table Question Answering" addresses the challenge of identifying relevant table content for question-answering tasks. The authors propose CABINET, a framework that enhances Large Language Models (LLMs) by suppressing irrelevant information and focusing on relevant parts of tables. CABINET consists of two main components: an Unsupervised Relevance Scorer (URS) and a Parsing Statement Generator. The URS is trained to assign relevance scores to table tokens based on their relevance to the input question, while the Parsing Statement Generator generates a natural language statement describing the criteria for relevant rows and columns. These components work together to weight and highlight relevant table content, improving the performance of LLMs in table QA tasks. The paper demonstrates that CABINET outperforms various baselines on three challenging datasets (WikiTQ, FeTaQA, and WikiSQL), is more robust to noise, and maintains high performance on larger tables. The authors also release their code and datasets to facilitate further research.The paper "CABINET: Content Relevance Based Noise Reduction for Table Question Answering" addresses the challenge of identifying relevant table content for question-answering tasks. The authors propose CABINET, a framework that enhances Large Language Models (LLMs) by suppressing irrelevant information and focusing on relevant parts of tables. CABINET consists of two main components: an Unsupervised Relevance Scorer (URS) and a Parsing Statement Generator. The URS is trained to assign relevance scores to table tokens based on their relevance to the input question, while the Parsing Statement Generator generates a natural language statement describing the criteria for relevant rows and columns. These components work together to weight and highlight relevant table content, improving the performance of LLMs in table QA tasks. The paper demonstrates that CABINET outperforms various baselines on three challenging datasets (WikiTQ, FeTaQA, and WikiSQL), is more robust to noise, and maintains high performance on larger tables. The authors also release their code and datasets to facilitate further research.