Measuring the Structural Similarity between Source Code Entities (S).

Abstract

Similarity coefficients are widely used in software engineering for several purposes, such as identification of refactoring opportunities and system remodularizations. Although the literature provides several similarity coefficients that vary on the computing strategy, there is a tendency among researchers to make habitual use of certain coefficients that others in their field are using. Consequently, some approaches might be using an inadequate coefficient for their purpose. In this paper, we conduct a quantitative study that compares 18 coefficients to identify which one is the most appropriate in determining where a class should be located. Our evaluation contemplates 111 open source systems from Qualitas Corpus, which totalizes more than 70,000 classes. As a result, we observed that Jaccard—one of the most used coefficients in our area—has not presented the best results. While Jaccard correctly indicated the suitable module to 22% of the classes, other coefficients were able to indicate 60%.