无线电通信技术2025，Vol.51Issue(3)：478-485,8.DOI:10.3969/j.issn.1003-3114.2025.03.006
基于结构化案例推理的SQL生成模型
SQL Generation Model Based on Structured Case Reasoning
摘要
Abstract
Automatic natural language-database interfaces are essential for application development.Inference time adaptive meth-od driven by semantic analysis can be compatible with samples from new domains without repeated tuning.Existing methods often sim-ply wire new domain input-output cases from encoder and input them into model,thus skew decoder.This approach does not utilize log-ical formal structure of the case.Based on above discussion,Structured Query Language(SQL)generation model based on Structured Case Reasoning Model(SCRM)is proposed,which uses subtree similarity between logical form of the case and candidate outputs to generate better decoder decisions.A trainable combinatorial subtree similarity function is proposed,which is more accurate to score large search boundaries than full-tree embedding.SCRM combines a new structured case memory and trainable query subtree similarity module to improve score of correct outputs during inference phase.Experiments are conducted on five different datasets,the proposed SCRM model achieves best results on different performance measurements with advantage of structured cases when compared to other models.关键词
Text-to-SQL/子树级相似性/结构化案例/解码器/相似函数Key words
Text-to-SQL/subtree-level similarity/structured case/decoder/similarity function分类
信息技术与安全科学引用本文复制引用
罗明全,王静,魏英..基于结构化案例推理的SQL生成模型[J].无线电通信技术,2025,51(3):478-485,8.基金项目
四川省哲学社会科学规划项目(SC22B141) Sichuan Provincial Philosophy and Social Science Planning Project(SC22B141) （SC22B141）
