Auto-weighted Multi-view learning for Semi-Supervised graph clustering

Abstract Despite the popularity of graph clustering, existing methods are haunted by two problems. One is the implicit assumption that all attributes are treated equally with the same weights. The other is that they treat graph as objects of pure topology, which cannot fully capture the structure of the graphs from each view. In this paper, to solve these problems, we first formulate the relationships between node attributes and various prior information as multi-view fusion. Then, by analyzing the modularity measure, a new edge function is devised, which provides a channel to allocate ideal weight for each attribute automatically and incorporate topology into graph clustering. Further, we propose a new similarity measure directly performing in the multi-view fusion, where not only the topology of a graph but also the additive information hidden in the graph is revealed. To verify the importance of different attributes from multi-view on clustering performance, we build an Auto-weighted Multi-view framework for Semi-Supervised (AMSS) graph clustering. Extensive experiments conducted on nine real-world datasets demonstrate the effectiveness and efficiency of the proposed method.