Building a tool for software code analysis a machine learning approach

Abstract

This article presents the application of a machine learning technique to a software code analysis tool. This tool builds a base of elementary and relevant program structures, acquired from the analysis of a primitive set of programs of good programming style. These program structures are compared to new programs to determine the quality of the latter. In this paper we stress the framework of the tool and discuss the critical details of its modules. The learning technique has been developed to use intensively a specific knowledge base in order to acquire the base of relevant program structures. We present problems that arise due to the necessity of using knowledge which is non-monotonic in nature. Particular issues will be highlighted by the analysis of requests written in the SQL language.