English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Graph contrastive learning (CL) methods learn node representations in aself-supervised manner by maximizing the similarity between the augmented noderepresentations obtained via a GNN-based encoder. However, CL methods performpoorly on graphs with heterophily, where connected nodes tend to belong todifferent classes. In this work, we address this problem by proposing aneffective graph CL method, namely HLCL, for learning graph representationsunder heterophily. HLCL first identifies a homophilic and a heterophilicsubgraph based on the cosine similarity of node features. It then uses alow-pass and a high-pass graph filter to aggregate representations of nodesconnected in the homophilic subgraph and differentiate representations of nodesin the heterophilic subgraph. The final node representations are learned bycontrasting both the augmented high-pass filtered views and the augmentedlow-pass filtered node views. Our extensive experiments show that HLCLoutperforms state-of-the-art graph CL methods on benchmark datasets withheterophily, as well as large-scale real-world graphs, by up to 7%, andoutperforms graph supervised learning methods on datasets with heterophily byup to 10%.

Graph Contrastive Learning under Heterophily via Graph Filters