Abstract:Graph-level representations are critical in various real-world applications, such as predicting the properties of molecules. However, in practice, precise graph annotations are generally very expensive and time-consuming. To address this issue, graph contrastive learning constructs an instance discrimination task, which pulls together positive pairs (augmentation pairs of the same graph) and pushes away negative pairs (augmentation pairs of different graphs) for unsupervised representation learning. However, since for a query, its negatives are uniformly sampled from all graphs, existing methods suffer from the critical sampling bias issue, i.e., the negatives likely having the same semantic structure with the query, leading to performance degradation. To mitigate this sampling bias issue, in this article, we propose a prototypical graph contrastive learning (PGCL) approach. Specifically, PGCL models the underlying semantic structure of the graph data via clustering semantically similar graphs into the same group and simultaneously encourages the clustering consistency for different augmentations of the same graph. Then, given a query, it performs negative sampling via drawing the graphs from those clusters that differ from the cluster of query, which ensures the semantic difference between query and its negative samples. Moreover, for a query, PGCL further reweights its negative samples based on the distance between their prototypes (cluster centroids) and the query prototype such that those negatives having moderate prototype distance enjoy relatively large weights. This reweighting strategy is proven to be more effective than uniform sampling. Experimental results on various graph benchmarks testify the advantages of our PGCL over state-of-the-art methods. The code is publicly available at https://github.com/ha-lins/PGCL.

Select Your Own Counterparts: Self-Supervised Graph Contrastive Learning With Positive Sampling

Debiased Graph Contrastive Learning.

ProGCL: Rethinking Hard Negative Mining in Graph Contrastive Learning

Improving Graph Contrastive Learning Via Adaptive Positive Sampling

B2-Sampling: Fusing Balanced and Biased Sampling for Graph Contrastive Learning

Enhancing Graph Contrastive Learning with Node Similarity

Boosting Graph Contrastive Learning Via Adaptive Sampling

Graph Self-Contrast Representation Learning

ConCur: Self-supervised Graph Representation Based on Contrastive Learning with Curriculum Negative Sampling

Group Contrastive Self-Supervised Learning on Graphs

SpeGCL: Self-supervised Graph Spectrum Contrastive Learning without Positive Samples

Generative Subgraph Contrast for Self-Supervised Graph Representation Learning

Prototypical Graph Contrastive Learning

Negative samples selecting strategy for graph contrastive learning

A Sample-driven Selection Framework: Towards Graph Contrastive Networks with Reinforcement Learning

Robust Hypergraph-Augmented Graph Contrastive Learning for Graph Self-Supervised Learning

A Negative Sample-Free Graph Contrastive Learning Algorithm

Contrastive Learning Meets Homophily: Two Birds with One Stone.

Select The Best: Enhancing Graph Representation with Adaptive Negative Sample Selection

Cluster-guided Contrastive Graph Clustering Network

GMCL: Graph Mask Contrastive Learning for Self-Supervised Graph Representation Learning