数据库查询中因果分析和解释的关键技术研究

由于现代信息技术的发展，对数据库查询处理的结果进行因果分析或解释将成为一种常态，其中因果分析和解释的视角分别是单个源元组和谓词。在对查询处理进行因果分析时，本项目通过编辑不等式连接查询溯源表达式，来计算其各个变量的弹性。因为基于谓词的解释采用了某种抽象，本项目拟进一步计算出输入数据的概要作为解释，并将提出高效的算法来满足这种需要。当查询的结果出现错误时，通常以检测输入数据为主，本项目希望通过查询日志来发现数据错误的根本原因和修复查询本身的错误。进一步当在事件流中监测到异常时，本项目拟提出算法来自动地把原始数据转换为特征空间，并采用基于熵的距离函数计算特征空间，然后把从特征值得来的解释进行优化建模处理。最后，研发一个数据库查询中因果分析和解释的原型系统，对本项目提出的理论和算法进行验证和分析。本项目的工作不仅对数据库查询中因果分析和解释有直接意义，而且对数据库和人工智能的有机结合有理论指导。

With the information technology development, it is normality to perform causal analysis and explanation of database queries. The perspectives of causality and explanation are single input tuple and predicate respectively. When performing causal analysis for the result tuple of a query, this project compiles the lineage of a conjunctive query with inequality and then computes the resilience of each variable in the lineage. Because it is based on some kind of abstract to use predicates for explanation, this project will compute the synopses of input data as explanation. This project will then introduce an efficiently algorithm to satisfy this requirement. When an error occurs in the query result, we often check the input data for error. Instead, this project will check the query log to find the errors in data and repair those errors. Further, if we monitor anomalies in event stream, this project will propose an algorithm to transform raw data into feature space in an automatic way. We use a new entropy-based distance function to calculate feature space and optimally model the explanation from the feature space. Finally, a causal analysis and explanation prototype is implemented based on the techniques developed in this project. This prototype is used to demonstrate our theories and algorithms. This project is not only useful for causal analysis and explanation for database queries but also for combining databases and artificial intelligence.