Community Detection in Attributed Networks via Kernel-Based Effective Resistance and Attribute Similarity

In attributed graphs, community detection methods group nodes by considering structural closeness and attribute similarity. In this work, we investigate the simultaneous use of kernels on the adjacency matrix of the graph and the node attribute similarity matrix. First, we weight the input adjacency matrix of the graph with the effective resistance between nodes, a Euclidean distance metric derived from the field of electric circuits that takes into account all the alternative paths between two nodes. Then we apply kernels for computing the similarity between nodes both in terms of structure and attributes. Simulations on synthetic networks show that kernels effectively improve the quality of the obtained partitions that better fit with the ground-truth.