Heterogeneous Graph Neural Network for Representation Learning

Explore a comprehensive lecture on Heterogeneous Graph Neural Networks presented by Chuxu Zhang at ICBS2024. Delve into the challenges and solutions for representation learning in heterogeneous graphs, focusing on incorporating both heterogeneous structural information and diverse node attributes. Learn about the innovative HetGNN model, which effectively combines heterogeneous graph data with various content types. Discover the model's architecture, including its random walk sampling strategy and dual-module neural network design for feature aggregation. Understand how HetGNN outperforms existing methods in tasks such as link prediction, recommendation, and node classification. This 50-minute talk provides valuable insights for researchers and practitioners working with complex, multi-modal graph data in various applications.