Pointwise Mutual Information-Induced Nonnegative Matrix Tri-Factorization With Hypergraph Regularization for Overlapping Community Detection

Overlapping community detection reveals nodes' affiliations, forming complex and dynamic intersecting structures in real-world networks. Currently, existing methods only consider decomposing the original adjacency matrix, which represents only the hard connections between nodes, ignoring potential links. Other methods use lower-order similarity graph, making it difficult to accomplish the task of overlapping community detection. To address these issues, this paper proposes a method based on the decomposition of the Pointwise Mutual Information matrix and hypergraphs. Specifically, the Pointwise Mutual Information (PMI) matrix is used instead of the traditional adjacency matrix to provide a more comprehensive view of interactions, effectively capturing both direct and indirect node interactions, thereby uncovering hidden community structures. Moreover, the hypergraph regularization naturally indicates high-order relationships among nodes, enabling more accurate detection of overlapping or nested communities. A Multiplicative Update Algorithm is introduced to solve the Hypergraph-based Nonnegative Matrix Tri-Factorization (NMTF) problem. Comprehensive experiments demonstrate that the proposed model holds significant advantages over other state-of-the-art models.