Query Answering for Tabular Data Using Large Language Models

This study presents a novel end-to-end approach to generating accurate responses for tabular data using Large Language Models combined with Retrieval-Augmented Generation. Our method leverages metadata augmentation and query expansion techniques to improve the accuracy of data matching and query response generation from a large pool of tabular datasets. Specifically, the system uses Large Language Models to enrich data descriptions and expand user queries, facilitating the retrieval of relevant data from approximately 70,000 tabular datasets. The proposed approach was demonstrated on five sample queries, each of which successfully generated correct answers. The results highlight the potential of Large Language Models to extract accurate statistical information from natural language input, providing a scalable solution for complex question-answering tasks involving structured data.