Conceptual Code Mining Mining for Source-Code Regularities with Formal Concept Analysis

Understanding the conceptual structure of large software systems, whether it is for software understanding or reengineering purposes, is a nontrivial task. In particular, knowing where to start the comprehension process is more difficult than it seems, especially when a system is large and complex and time is scarce. We propose an approach to mine a system’s source code automatically and efficiently for relevant concepts of interest, which we refer to as source-code regularities: what concerns are addressed in the code, what patterns, programming idioms and conventions have been adopted, and where and how they are implemented. We use formal concept analysis to do the actual source-code mining, and then filter, classify and combine the results to present them in a format that is more convenient to a software engineer. We applied a prototype tool that implements this approach to several small to medium-sized Smalltalk applications. For each of these, the tool discovered several interesting source-code regularities. Although the tool and approach can still be improved in many ways, the tool does already provide useful results when having a first contact with a system. The obtained results also illustrate the relevance and feasibility of using formal concept analysis as a technique for source code mining.