Refining Similarity Matrices for Clustering Attributed Networks Accurately

Abstract

Due to the recent trend of the Social Network and the increase in the number of academic papers published in the world, attributed networks consisting of relationships between objects such as humans and the papers are becoming huge. Therefore, various studies for clustering the attributed networks into some sub-networks are actively conducted. When clustering the attributed networks with the spectral clustering, the accuracy of the spectral clustering is greatly affected by quality of similarity matrices representing similarities between the objects. In this study, we aim to improve the accuracy by refining the matrices before applying the matrices to the spectral clustering. Furthermore, we verify practicability of our proposed method comparing accuracies of the spectral clustering with similarity matrices before and after refining.