Reproduce “TableZoomer: A Collaborative Agent Framework for Large-scale Table Question Answering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sishi Xiong, Ziyang He, Zhongjiang He, Yu Zhao, Changzai Pan, Jie Zhang, Zhenhe Wu, Shuangyong Song and 1 more
Developed by TeleAI, TableZoomer is an LLM-powered agent framework designed for large-scale Table Question Answering, efficiently addressing the challenges of massive and heterogeneous tables. It achieved absolute accuracy improvements of 19.34% on DataBench by employing schema-based representation and query-aware zooming, significantly reducing input tokens while maintaining high performance.
Key insights
- Replacing full table representations with structured, semantically enriched table schemas significantly reduces token usage (from O(M × N) to O(N)) and enhances LLM focus on critical metadata.
- Dynamically zooming into query-relevant sub-tables through column selection and entity linking is essential for efficient information localization in large tables, mitigating context window limitations.
- A synergistic approach of programming-based reasoning (Program-of-Thoughts) coupled with an iterative ReAct paradigm offers robust numerical computation and self-correction, effectively countering LLM-specific reasoning flaws like hallucination.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.