Community Detection in Networks using PageRank Contributions

The modern science of networks has made significant contributions to our understanding of complex real world systems. One of the most relevant features of graphs representing the real world networks is their community structure. Therefore, a large body of work in industry and academia has been devoted to identifying community structure in complex networks. In this thesis, we design PC-KM, a variation of $k$-means clustering using PageRank contributions to detect communities in networks. In order to scale to large size networks, we propose another method PPC-KM, which uses random projections to reduce dimensionality while preserving features required for community detection. We also present a fuzzy version of PPC-KM to consider the overlapping communities in networks. We evaluate our algorithms on several datasets. The results show that our methods detect communities with high performance on real world networks.