SDAM: Structure-Difference-Aware Memory Evolution for Complex Text-to-SQL

arXiv:2608.123382026-08-14

arXiv:2608.12338v1 Announce Type: new Abstract: Text-to-SQL aims to convert natural language questions into executable SQL queries. While memory-based agent system improves complex SQL generation, existing memory design neglect historical experience and suffer from weak structure analysis, shallow semantic understanding, and poor schema alignment. To address these challenges, we propose SDAM. Specifically, SDAM identifies potential errors via a structure-difference aware reasoning tree, extracts

저자 · Keyan Xu, Dingzirui Wang, Xuanliang Zhang, Qingfu Zhu, Wanxiang Che