Local expansion and optimization for higher-order graph clustering

Graph clustering aims to identify clusters that feature tighter connections between internal nodes than external nodes. We noted that conventional clustering approaches based on a single vertex or edge cannot meet the requirements of clustering in a higher-order mixed structure formed by multiple nodes in a complex network. Considering the above limitation, we are aware of the fact that a clustering coefficient can measure the degree to which nodes in a graph tend to cluster, even if only a small area of the graph is given. In this study, we introduce a new cluster quality score, i.e., the local motif rate, which can effectively respond to the density of clusters in a higher-order graph. We also propose a motif-based local expansion and optimization algorithm (MLEO) to improve local higher-order graph clustering. This algorithm is a purely local algorithm and can be applied directly to higher-order graphs without conversion to a weighted graph, thus avoiding distortion of the transform. In addition, we propose a new seed-processing strategy in a higher-order graph. The experimental results show that our proposed strategy can achieve better performance than the existing approaches when using a quadrangle as the motif in the LFR network and the value of the mixing parameter μ exceeds 0.6.