基于NL2SQL的兵棋数据智能统计分析方法研究

Research on Intelligent Statistical Analysis of Wargaming Data Based on NL2SQL

面对海量的兵棋数据,传统界面查询的方式已经无法满足指挥员快速、全面、精准查询数据的要求。通过深入分析兵棋数据特点与主流NL2SQL(natural language to struct query language)模型的缺陷,提出了一套适合兵棋数据智能统计查询的解决方案。针对领域数据集缺乏,提出了一套基于人机协助、动态迭代的兵棋数据集构建方案;针对兵棋查询问句时间敏感的问题,提出了一套“规则+深度学习”的时间表达式识别与规范方法;针对兵棋数据量大提取查询值困难的问题,修改完善了Bridge模型的值提取与SQL生成架构。综合运用以上方案,使兵棋数据查询的精准匹配准确率达到75%以上。

In the face of massive wargaming data,the traditional interface query method can no longer meet the commander's requirements,i.e.,fast,comprehensive,and accurate data querying.Through indepth analysis of the characteristics of wargaming data and the defects of the mainstream natural language to struct query language(NL2SQL)model,a set of solutions for the intelligent statistical query of wargaming data is presented.Due to the lack of datasets,a wargaming dataset construction scheme based on human-machine assistance and dynamic iteration is provided.In order to solve the timesensitive problem of wargaming querying,time expression recognition and standardization methods based on"rule+deep learning"are proposed.The value extraction and SQL generation architecture of the Bridge model are modified to facilitate the extraction of query value for a large amount of wargaming data.By comprehensively using the above scheme,the query accuracy of wargaming data is significantly enhanced to more than 75%.