MURRE: Multi-Hop Table Retrieval with Removal for Open-Domain Text-to-SQL

  • 2024-09-18 03:48:25
  • Xuanliang Zhang, Dingzirui Wang, Longxu Dou, Qingfu Zhu, Wanxiang Che
  • 0

Abstract

The open-domain text-to-SQL task aims to retrieve question-relevant tablesfrom massive databases and generate SQL. However, the performance of currentmethods is constrained by single-hop retrieval, and existing multi-hopretrieval of open-domain question answering is not directly applicable due tothe tendency to retrieve tables similar to the retrieved ones but irrelevant tothe question. Since the questions in text-to-SQL usually contain all requiredinformation, while previous multi-hop retrieval supplements the questions withretrieved documents. Therefore, we propose the multi-hop table retrieval withremoval (MURRE), which removes previously retrieved information from thequestion to guide the retriever towards unretrieved relevant tables. Ourexperiments on two open-domain text-to-SQL datasets demonstrate an averageimprovement of 5.7% over the previous state-of-the-art results.

 

Quick Read (beta)

loading the full paper ...