Improved NL2SQL method based on generative large language model

• Main Authors: XU JIWEI, DUAN CHUNXIAN, FU ZHUO, HAN XIAOLE, LI XIAOCHAO, CHEN SHENGPENG, LEI ZHEN, LIU MENGJUN, XIA WEI, WANG JINGPEI, LI YING, WANG FENG, LIU GAO
• Format: Patent
• Language: Chinese; English
• Published: 29.09.2023
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention is suitable for the technical field of natural language processing, and provides an improved NL2SQL method based on a generative large language model, which comprises the following steps: S1, preprocessing table information of a database; s2, preprocessing the natural language question sentences; s3, matching a target table; s4, outputting a large language model result; and S5, extracting the SQL statement and outputting the SQL statement. According to the method, the generative large language model is used, the prompt statement is constructed in a thinking chain mode to improve the matching precision, end-to-end retraining is not needed, and the method is better in applicability in an actual production environment in which a database table structure is frequently updated. Besides, the database query semantic recognition problem under the multi-table repeated column interference environment in the actual production environment is solved by adopting a mode of pre-calculating table and column weig