Noori, AzadBalafar, Mohammad AliBouyer, AsgaraliSalmani, Khosro2025-04-172025-04-172025Noori, A., Balafar, M. A., Bouyer, A., & Salmani, K. (2025). A novel contrastive multi-view framework for heterogeneous graph embedding. Knowledge and Information Systems, 1-27.0219-13770219-3116http://dx.doi.org/10.1007/s10115-025-02346-0https://hdl.handle.net/20.500.12713/6304Heterogeneous graphs, characterized by diverse node types and relational structures, serve as powerful tools for representing intricate real-world systems. Understanding the complex relationships within these graphs is essential for various downstream tasks. However, traditional graph embedding methods often struggle to represent the rich semantics and heterogeneity of these graphs effectively. In order to tackle this challenge, we present an innovative self-supervised multi-view network (SMHGNN) designed for the embedding of heterogeneous graphs. SMHGNN utilizes three complementary views: meta-paths, meta-structures, and the network schema, to thoroughly capture the complex relationships and interactions among nodes in heterogeneous graphs. The proposed SMHGNN utilizes a self-supervised learning paradigm, thereby obviating the necessity for annotated data while simultaneously augmenting the model's capability for generalization. Furthermore, we present an innovative mechanism for the identification of positive and negative samples, which is predicated on a scoring matrix that integrates both node characteristics and graph topology, thereby proficiently differentiating authentic positive pairs from false negatives. Comprehensive empirical evaluations conducted on established benchmark datasets elucidate the enhanced efficacy of our SMHGNN in comparison with cutting-edge methodologies in heterogeneous graph embedding. (The proposed method achieved a 0.75-2.4% improvement in node classification and a 0.78-2.7% improvement in clustering across diverse datasets when compared to previous state-of-the-art methods.).eninfo:eu-repo/semantics/closedAccessGraph Neural NetworksHeterogeneous Graph EmbeddingMeta-PathsMeta-StructuresScore MatrixSelf-Supervised LearningArticleWOS:0014105829000012-s2.0-85217381254Q210.1007/s10115-025-02346-0Q2