Applying Complex Network Method to Software Clustering

There has been considerable interest in network motif for understanding network local features, growth and evolution mechanisms. In order to discover the relationship between software networks and various realistic networks and apply complex network community detection methods to software clustering, we extended the network motif research to software domain. After comparing triad significance profiles from 138 Java open source software packages, we found that software networks could be divided into 3 clusters which are consistent with the known super-families from various other types of networks. It seems that software scale may be one of the reasons causing different motif SP distribution. Most of middle and large scale software networks have similar local structure with biological networks. They may share the same design and evolving principles. Moreover, we applied the community detection algorithm of complex networks to the software clustering problem and made comparisons with bunch using the same clustering criterion. The results of our experiment show that the clustering result is better than the bunch method.