Clustering Source Code Elements by Semantic Similarity Using Wikipedia

For humans it is no problem to determine if two words have a high or low semantic similarity in a given context. But is it possible to support a software developer or architect by using semantic data extracted from source code in the same way other relations like typical source code relations do? To answer this question we developed an approach to compute the semantic similarity by using Wikipedia as a textual corpus. In a case study we demonstrate this approach with a manageable software system. The results of using semantic similarities are compared with the outcome of using source code relations instead.