信息工程学院

讲座时间：2023年11月24日13时00分

讲座地点：工B105

讲座对象：信息工程学院师生

讲座摘要：

Representation learning in heterogeneous graphs is challenging due to their complexity. Current self-supervised learning (SSL) methods mainly focus on node-level tasks, neglecting global graph features. They also use computationally demanding techniques and overlook certain node relationships, reducing graph-level learning effectiveness. To address these issues, we introduce a novel SSL framework for heterogeneous hypergraph learning. This framework uses a k-hop neighborhood method and a shared attribute system instead of traditional meta-path techniques. We also offer a context-aware pooling system and a high-order-aware augmentation strategy, improving graph-level learning. Tested against various baselines, our model shows a 5.81% performance boost.