Combining Latent Semantic Indexing and Call Graphs to Improve Feature Location

P. Shao, N.A. Kraft, and R.K. Smith (USA)


Information Retrieval, Latent Semantic Indexing, Call Graph, Feature Location


Best software engineering practices encourage developers to use descriptive labels in their source code. These labels include meaningful names for classes, methods, functions, and variables. When performing maintenance or refactoring tasks, these labels are treated as features, and the task to locate these free text terms in source code is commonly accomplished using information retrieval (IR) techniques. Existing IR-based feature location techniques consider lexical information but ignore structural information. In this paper, we propose a new feature location technique, combines latent semantic indexing (LSI), an IR technique that operates on lexical information, and call graphs, which encode structural information, to provide better accuracy than using LSI alone. We then present a case study in which we apply our technique to three open source Java projects. The results demonstrate that combining LSI and call graph improves the accuracy of feature location in source code when compared to using LSI alone.

Important Links:

Go Back