Graph Clustering Based on Attribute-Aware Graph Embedding

Graph clustering is a fundamental problem in graph mining and network analysis. To group vertices of a graph into a series of densely knitted clusters with each cluster being well-separated from all the others, classic methods primarily consider the mere graph structure information in modeling and quantifying the proximity or distance of vertices for graph clustering. However, with the proliferation of rich, heterogeneous attribute information widely available in real-world graphs, such as user profiles in social networks, and GO (Gene Ontology) terms in protein interaction networks, it becomes essential to combine both structure and attribute information of graphs towards yielding better-quality clusters. In this chapter, we propose a new graph embedding approach for attributed graph clustering. We embed each vertex of a graph into a continuous vector space within which the local structure and attribute information surrounding the vertex can be jointly encoded in a unified, latent representation. Specifically, we quantify the vertex-wise attribute proximity into edge weights and leverage a group of truncated, attribute-aware random walks to learn the latent representations of vertices. This way, the challenging attributed graph clustering problem can be cast into the traditional problem of multidimensional data clustering, which has admitted efficient and cost-effective solutions. We apply our attribute-aware graph embedding algorithm in a series of real-world and synthetic attributed graphs and networks. The experimental studies demonstrate that our proposed method significantly outperforms the state-of-the-art attributed graph clustering techniques in terms of both clustering effectiveness and efficiency.