Reproduce “ST-Raptor: LLM-Powered Semi-Structured 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
Zirui Tang, Boyu Niu, Xuanhe Zhou, Boxiu Li, Wei Zhou, Jiannan Wang, Guoliang Li, Xinyi Zhang and 1 more
ST-Raptor introduces a Hierarchical Orthogonal Tree (HO-Tree) model and a pipeline-based question answering framework to process semi-structured tables. It achieved 72.39% accuracy on the new SSTQA benchmark, outperforming existing baselines by over 10.23%.
Key insights
- Explicitly modeling 2D table structure with the HO-Tree preserves critical layout information, enabling more accurate and robust reasoning by LLMs.
- Decomposing complex questions into atomic operations and leveraging bi-directional traversal strategies within the HO-Tree facilitates multi-hop and multi-directional reasoning.
- Robust forward and backward verification stages are crucial for mitigating LLM hallucination and enhancing the reliability of answers for complex tabular QA tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.