Mining Source Code Regularities with FCA and Association Rules

摘要

Software encodes knowledge about the application domain as well as about the solution domain, i.e. about the used technologies and how they are combined. The correct implementation at this level is very often characterized by regularities, which are seldom expressed explicitly. Therefore it is useful to identify these regularities automatically. To name two obvious examples: "95% of the classes that implement equals () implement as well hashCode ()" or "80% of the methods that open a transaction close it as well". The deviations from these rules are at least worth a review if not implementation errors. Formal Concept Analysis offers a formal framework to mine Association Rules automatically and Concept Lattices are helpful in gaining an intuitive understanding.