Robust heterogeneous network representation learning by multifaceted curriculum training.

The use of heterogeneous networks has gained significant traction for modeling and analyzing complex systems across diverse domains because of their ability to represent various types of entities and relationships. However, these networks face considerable challenges in representation due to different types of noise, including node feature noise, edge noise, and label noise, which arise from data collection imperfections, inconsistent labeling processes, and network construction errors. Despite the rich literature on curriculum learning (CL), existing approaches have not sufficiently integrated or motivated its application in heterogeneous networks. In this paper, we address these challenges by leveraging network structures and investigating the integration of Curriculum Learning (CL) to enhance the robustness of GNNs against multiple forms of noise in heterogeneous networks to obtain precise representations. We propose a novel approach, MultifaceteD CurricuLum (MDCL), which adaptively incorporates multifaceted measures to capture various aspects of heterogeneous networks, including node features, topological structures, and training dynamics. MDCL utilizes an adaptive weighting mechanism to make dynamic decisions on difficulty prioritization, thereby simulating a robust representation learning process in the presence of complex noise. Empirical evaluations on benchmark datasets and GNN architectures demonstrate that MDCL consistently improves the accuracy and robustness of GNNs in scenarios with diverse noise types, establishing it as a promising solution for real-world applications involving heterogeneous networks.