Analyzing Term Weighting Schemes for Labeling Software Clusters

Clustering techniques have been widely employed for software modularization. The clusters formed as a result of the clustering process may be difficult to understand unless they are appropriately labeled. One method to assign labels is to use term weighting schemes from Information Retrieval and Text Categorization which use weights to assign importance to terms in a document. Some of these term weighting schemes have been used by researchers for labeling clusters, but there is a need to compare various schemes and analyze their strengths and weaknesses. In this paper, we analyze four different schemes in the context of software and identify cases where one may be better than the other. We also conduct experiments to verify the behavior of the weighting schemes according to software characteristics.