Sharpening deep graph clustering via diverse bellwethers

Peiyao Zhao, Xin Li, Yuangang Pan, Ivor W. Tsang, Mingzhong Wang, Lejian Liao

Abstract: Highlights•Existing deep graph clustering methods are lack of adequate support for learning diverse clusters.•A diversity regularizer term is proposed for the gradient function of cluster centroids.•A ME-based clustering loss is proposed to sharpen the clustering assignment distributions.•Experiments verify the effectiveness of each technique and enhancing node discrimination.