Reproduce “HybridQA: A Dataset of Multi-Hop Question Answering over Tabular and Textual Data” 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

Wenhu Chen, Hanwen Zha, Zhiyu Chen, Wenhan Xiong, Hong Wang, William Wang

HybridQA pairs Wikipedia tables with linked passages in a question-answering dataset designed for questions requiring both sources. HYBRIDER connects evidence at the table-cell level, ranks candidate cells, moves between related cells, and either copies cell text or reads linked passages for an answer. On the test set, its BERT-base model reaches 42.5% exact match, versus 8.3% for Table-Only and 19.1% for Passage-Only; this comparison is specific to the benchmark and these model pipelines.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.