PIPER: Content-Based Table Search via profiling and LLM-Generated Pseudoqueries

In three linesPIPER is a content-driven retrieval method for tabular datasets using table profiles and LLM-generated pseudoqueries for dense retrieval. It outperforms metadata-based baselines and existing TableQA methods in poor-metadata settings.

Summary generated by Claude — human-verified