Labeling Feature-Oriented Software Clusters for Software Visualization Application

Software clustering techniques have been used to analyze the reality of software structure. The visualization of the detected clusters has also been studied. However, the features implemented by the detected clusters are not obvious and understanding them is a crucial part of the industrial use of software clustering. In this study, we examined the existing information retrieval method and found three major issues it has. We developed technical solutions for each of them: using hierarchical labeling, weighing the words likely representing the feature by considering an architectural convention, and modifying the idf score by the scale of the cluster. The effectiveness of our approach is validated through case studies using actual software products including a COBOL business application. Also, we faced two additional practical problems: effectiveness of the method words and plural and conjugated forms of the words. We found the method name words were less useful than the class name words, and lemmatization was successfully used to normalize the form of the words even in the case of program identifiers.