Analysis of Text to SQL Method Based on Relational T5 and Reordering Decoding
This paper expounds that a relational T5 model with a re-ranking decoding module is constructed for the Text-to-SQL task.Specifically,introduces a relationship-aware self-attentive module into the T5 encoder as a way to improve the model's semantic understanding of various relationship structures.In addition,in the decoding process,this paper introduces re-ranking decoding to the N-best list obtained by PICARD,which can be used to check whether the entity name in each prediction in the list appears in the database by a search algorithm to improve the correctness of the selection of results.
NETL
NSTL
万方数据
