Deep Graph Clustering in Social Network

Pengwei Hu, Keith C. C. Chan, Tiantian He

Abstract: In this paper, we present deep attributes residue graph algorithm (DARG), a novel model for learning deep representations of graph. The algorithm can discover clusters by taking into consideration node relevance. DARG does so by first learns attributes relevance and cluster deep representations of vertices appearing in a graph, unlike existing work, integrates content interactions of the nodes into the graph learning process. First, the relevance of contents between each node pair within the network is abstracted. Then we turn the problem of computing the first k eigenvectors in spectral clustering into a computing deep representations task. This model just need learns content information to represent vertices appearing in a graph and without the need for considering topological information. Such content information is much easier to obtain than topological links in the real world. We conduct an experiment on SNAP Facebook dataset, empirical results demonstrate that proposed approach significantly outperforms other state-of-the-art methods in such task.