Three-Way Decisions Community Detection Model Based on Weighted Graph Representation

Abstract

Community detection is of great significance to the study of complex networks. Community detection algorithm based on three-way decisions (TWD) forms a multi-layered community structure by hierarchical clustering and then selects a suitable layer as the community detection result. However, this layer usually contains overlapping communities. Based on the idea of TWD, we define the overlapping part in the communities as boundary region (BND), and the non-overlapping part as positive region (POS) or negative region (NEG). How to correctly divide the nodes in the BND into the POS or NEG is a challenge for three-way decisions community detection. The general methods to deal with boundary region are modularity increment and similarity calculation. But these methods only take advantage of the local features of the network, without considering the information of the divided communities and the similarity of the global structure. Therefore, in this paper, we propose a method for three-way decisions community detection based on weighted graph representation (WGR-TWD). The weighted graph representation (WGR) can well transform the global structure into vector representation and make the two nodes in the boundary region more similar by using frequency of appearing in the same community as the weight. Firstly, the multi-layered community structure is constructed by hierarchical clustering. The target layer is selected according to the extended modularity value of each layer. Secondly, all nodes are converted into vectors by WGR. Finally, the nodes in the BND are divided into the POS or NEG based on cosine similarity. Experiments on real-world networks demonstrate that WGR-TWD is effective for community detection in networks compared with the state-of-the-art algorithms.