Meta-path and context-aware learning for attribute completion in heterogeneous graphs.

Abstract

Heterogeneous graph attribute completion (HGAC) has garnered substantial attention due to its critical role in various graph-based applications in recent years. However, existing HGAC methods often overlook valuable meta-path information, focus solely on neighboring nodes during attribute completion, and exhibit inefficiencies when integrated with heterogeneous graph neural networks for downstream tasks. To address these challenges, we propose EMC-Net, a novel learning framework that incorporates meta-path information and context-aware learning. Our framework employs collaborative meta-path-driven embedding schemes, which capture the rich semantic information embedded in heterogeneous graphs, and introduce context-aware attention mechanisms to dynamically adjust the importance of different nodes and edges during attribute completion. Our framework not only enhances the accuracy of attribute predictions but also improves the efficiency and scalability of the model. Extensive experimental results on benchmark datasets demonstrate that our framework significantly outperforms existing HGAC methods in terms of both accuracy and computational efficiency. This advancement offers new insights and methodologies for developing more robust and effective heterogeneous graph attribute completion technologies.