On October 9, Madelon Hulsebos, a researcher and former BIDS-Accenture Data Science Research Scholar, presented her work “Retrieval Systems for Structured Data: the critical missing piece for grounding LLM-driven query interfaces in factual data”. She discussed how retrieval systems for structured data are key in grounding LLM-powered query interfaces in factual and domain-specific data stored in relational databases and data lakes.
The first part of Madelon's talk focused on helping data analysts and machine learning engineers find the right dataset for deeper analysis. In the second part, she looked at how to enrich large language model interfaces with structured data and how to ground LLM-powered interfaces in structured data. She concluded by introducing TARGET, a benchmark for evaluating TAble Retrieval for GEnerative Tasks. TARGET is used to analyze the retrieval performance of different retrievers in isolation, as well as their impact on downstream generators for question answering, fact verification, and text-to-SQL.
