Semi-supervised Co-Clustering on Attributed Heterogeneous Information Networks

•Designing a learnable overall relevance measure that makes full use of structural and attributed information by integrating HeteSim and attribute projection.•Proposing a constrained negative matrix tri-factorization which utilizes pairwise constraints to cluster nodes of different types at the same time and give a closed solution.•Modeling a unified framework to simultaneously co-cluster different-type nodes and mine the latent relevance between heterogeneous clusters on realworld attributed HINs such as e-commerce platforms and bibliographic networks.