High-order quasi-clique detection based on motif similarity for complex networks

Dense subgraph mining is pivotal for uncovering cohesive modules in complex networks, and quasi-cliques provide a practical relaxation of strict cliques. However, existing methods primarily rely on local pairwise connectivity, which may overlook higher-order patterns, such as motifs and their derived instances. To address this limitation, this study introduces a motif-based framework that leverages higher-order structural similarity to identify cohesive vertex groups. First, we enhance the Neighborhood-Based Similarity (NBSim) algorithm by introducing a node-adaptive threshold based on local clustering and motif participation, thereby improving the density of detected quasi-cliques. We then propose the Motif-Based Similarity (MBSim) algorithm, an expansion-based algorithm that selects neighbors based on motif similarity to detect denser subgraphs with competitive scale. Experiments using the proposed method on the World Trade Network (WTN) from 2000 to 2019 reveal a core–periphery trade pattern and a disparity between structural and functional robustness.