Clustering in Networks with Multi-Modality Attributes

Network clustering is one of the most significant tasks of network analytics. To discover network clusters, there have been many approaches proposed, utilizing network topology, or node attributes. However, there are no effective approaches that are able to discover clusters in the network with multiple modalities of attributes. In this paper, we propose a novel clustering model, called CNMMA, to discover network clusters using edge structure, and multi-modality attributes associated with vertices. Assuming edge structure, and node attributes are generated by corresponding low dimensional latent spaces (matrices), CNMMA can learn an optimal latent matrix representing the cluster membership for each vertex in the network. Besides, CNMMA makes use of an effective method to regulate the latent spaces w.r.t. edge structure and node attributes so that those vertices sharing similar edges and modality-wise attributes are more possible to be assigned with the same cluster labels. CNMMA has been tested with several real-world networks, which contain multiple modalities of node attributes, and has been compared with state-of-the-art approaches to network clustering. The experimental results show that CNMMA outperforms most approaches in most datasets. The clusters discovered by CNMMA are better matched with the ground truth.