Abstract:Although graph neural networks (GNNs) have achieved impressive achievements in graph classification, they often need abundant task-specific labels, which could be extensively costly to acquire. A credible solution is to explore additional labeled graphs to enhance unsupervised learning on the target domain. However, how to apply GNNs to domain adaptation remains unsolved owing to the insufficient exploration of graph topology and the significant domain discrepancy. In this paper, we propose Coupled Contrastive Graph Representation Learning (CoCo), which extracts the topological information from coupled learning branches and reduces the domain discrepancy with coupled contrastive learning. CoCo contains a graph convolutional network branch and a hierarchical graph kernel network branch, which explore graph topology in implicit and explicit manners. Besides, we incorporate coupled branches into a holistic multi-view contrastive learning framework, which not only incorporates graph representations learned from complementary views for enhanced understanding, but also encourages the similarity between cross-domain example pairs with the same semantics for domain alignment. Extensive experiments on popular datasets show that our CoCo outperforms these competing baselines in different settings generally.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the graph classification task of unsupervised domain adaptation in graph - structured data. Specifically, although graph neural networks (GNNs) have made remarkable achievements in graph classification tasks, they usually require a large number of task - specific labels, which are often very expensive or even unavailable. In addition, due to the insufficient extraction of graph topological information and the significant existence of inter - domain differences, the existing methods perform poorly when dealing with graph data between different domains. Therefore, this research aims to explore how to use additional labeled graphs to enhance unsupervised learning on the target domain while reducing the differences between different domains to improve the performance of graph classification. To address the above challenges, the authors propose a coupled contrastive framework (Coupled Contrastive Graph Representation Learning, CoCo). This framework extracts the topological information of graphs through two complementary branches (i.e., the graph convolutional network branch and the hierarchical graph kernel network branch), and reduces the inter - domain differences through a multi - view contrastive learning framework, thereby achieving effective unsupervised domain - adapted graph classification. The specific contributions are as follows: 1. **Propose a new unsupervised domain - adapted graph classification method**: named CoCo, which contains a graph convolutional network branch and a hierarchical graph kernel network branch to mine the topological information of graphs from different perspectives. 2. **Cross - branch contrastive learning**: Encourage the two branches to reach an agreement on each graph sample and generate high - quality graph representations containing comprehensive structural information. 3. **Cross - domain contrastive learning**: Effectively reduce the inter - domain differences by minimizing the distance between cross - domain sample pairs with the same semantics. 4. **Experimental verification**: A large number of experiments have been carried out on multiple widely recognized graph classification benchmark datasets, and the results show that the proposed CoCo significantly outperforms a series of competing baseline methods. Through the above methods, CoCo can not only effectively extract the structural information of graphs from limited labeled data, but also reduce the differences between different domains through cross - domain contrastive learning, thereby improving the performance of graph classification.

CoCo: A Coupled Contrastive Framework for Unsupervised Domain Adaptive Graph Classification

DiscoGNN: A Sample-Efficient Framework for Self-Supervised Graph Representation Learning

CoCoS: Enhancing Semi-supervised Learning on Graphs with Unlabeled Data via Contrastive Context Sharing

GraphCoCo: Graph Complementary Contrastive Learning

Graph Communal Contrastive Learning

NNC-GCN: Neighbours-to-Neighbours Contrastive Graph Convolutional Network for Semi-Supervised Classification

Self-supervised Heterogeneous Graph Neural Network with Co-contrastive Learning

Hierarchical Contrastive Learning Enhanced Heterogeneous Graph Neural Network

CoLM2S: Contrastive Self‐supervised Learning on Attributed Multiplex Graph Network with Multi‐scale Information

CuCo: Graph Representation with Curriculum Contrastive Learning

Graph Representation Learning via Contrasting Cluster Assignments

Domain Adaptive Graph Classification

Mixed Graph Contrastive Network for Semi-Supervised Node Classification

Semi-Supervised Mixture Learning for Graph Neural Networks With Neighbor Dependence

Deep Graph Contrastive Representation Learning

GTC: GNN-Transformer co-contrastive learning for self-supervised heterogeneous graph representation

Dynamic Graph Convolutional Networks by Semi-Supervised Contrastive Learning

CC-GNN: A Clustering Contrastive Learning Network for Graph Semi-Supervised Learning

Adaptive Multi-layer Contrastive Graph Neural Networks