Abstract:Heterogeneous graph neural networks (HGNNs) as an emerging technique have shown superior capacity of dealing with heterogeneous information network (HIN). However, most HGNNs follow a semi-supervised learning manner, which notably limits their wide use in reality since labels are usually scarce in real applications. Recently, contrastive learning, a self-supervised method, becomes one of the most exciting learning paradigms and shows great potential when there are no labels. In this paper, we study the problem of self-supervised HGNNs and propose a novel co-contrastive learning mechanism for HGNNs, named HeCo. Different from traditional contrastive learning which only focuses on contrasting positive and negative samples, HeCo employs cross-view contrastive mechanism. Specifically, two views of a HIN (network schema and meta-path views) are proposed to learn node embeddings, so as to capture both of local and high-order structures simultaneously. Then the cross-view contrastive learning, as well as a view mask mechanism, is proposed, which is able to extract the positive and negative embeddings from two views. This enables the two views to collaboratively supervise each other and finally learn high-level node embeddings. Moreover, to further boost the performance of HeCo, two additional methods are designed to generate harder negative samples with high quality. The essence of HeCo is to make positive samples from different views close to each other by cross-view contrast, and learn the factors invariant to two proposed views. However, besides the invariant factors, view-specific factors complementally provide the diverse structure information between different nodes, which also should be contained into the final embeddings. Therefore, we need to further explore each view independently and propose a modified model, called HeCo++. Specifically, HeCo++ conducts hierarchical contrastive learning, including cross-view and intra-view contrasts, which aims to enhance the mining of respective structures. Extensive experiments conducted on a variety of real-world networks show the superior performance of the proposed methods over the state-of-the-arts.

Hypergraph contrastive attention networks for hyperedge prediction with negative samples evaluation

ProGCL: Rethinking Hard Negative Mining in Graph Contrastive Learning

Debiased Graph Contrastive Learning.

AHP: Learning to Negative Sample for Hyperedge Prediction

Aggregative and Contrastive Dual-View Graph Attention Network for Hyperspectral Image Classification

Negative samples selecting strategy for graph contrastive learning

Negative Samples Mining Matters: Reconsidering Hyperspectral Image Classification with Contrastive Learning

Hard Sample Aware Network for Contrastive Deep Graph Clustering

Knowledge Graph Completion Based on Hyperbolic Graph Contrastive Attention Network

Ha-gnn: a novel graph neural network based on hyperbolic attention

Scalable and Effective Negative Sample Generation for Hyperedge Prediction

Hierarchical Contrastive Learning Enhanced Heterogeneous Graph Neural Network

Domain-adaptive Graph Attention-supervised Network for Cross-network Edge Classification

A Contrastive Knowledge Graph Embedding Model with Hierarchical Attention and Dynamic Completion

CAT: A Causally Graph Attention Network for Trimming Heterophilic Graph

CAHAN: Drug-Disease Association Prediction Based on Cross-Attention Mechanism

Enhancing Hyperedge Prediction with Context-Aware Self-Supervised Learning

Cluster-guided Contrastive Graph Clustering Network

Generating Counterfactual Hard Negative Samples for Graph Contrastive Learning

Biomedical Interaction Prediction with Adaptive Line Graph Contrastive Learning