Abstract | ||
---|---|---|
Graph Convolutional Networks (GCNs) have been proven to be effective in various graph-related tasks, such as community detection. Essentially graph convolution is simply a special form of Laplacian smoothing, acting as a low-pass filter that makes the features of nodes linked to each other similarly. For community detection, however, the similarity of intra-community nodes and the difference of inter-community nodes are equally vital. To bridge the gap between GCNs and community detection, we develop a novel Community-Centric Dual Filter (CCDF) framework for community detection. The central idea is that, besides of low-pass filter in GCN, we define network modularity enhanced high-pass filter to separate the discriminative signals from the raw features. In addition, we design a scheme to jointly optimize low-frequency and high-frequency information extraction on statistical modeling of Markov Random Fields. Extensive experiments demonstrate that the proposed CCDF model can consistently outperform or match state-of-the-art baselines in terms of semi-supervised community detection. |
Year | DOI | Venue |
---|---|---|
2021 | 10.1007/978-3-030-82153-1_15 | KNOWLEDGE SCIENCE, ENGINEERING AND MANAGEMENT, PT III |
Keywords | DocType | Volume |
Graph Convolutional Networks, Community detection, Modularity, Markov Random Field | Conference | 12817 |
ISSN | Citations | PageRank |
0302-9743 | 0 | 0.34 |
References | Authors | |
0 | 4 |