{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/new-benchmarks-for-learning-on-non","title":"New Benchmarks for Learning on Non-Homophilous Graphs","arxiv_id":"2104.01404","date":"2021-04-03","proceeding":null,"authors":["Derek Lim","Xiuyu Li","Felix Hohne","Ser-Nam Lim"],"abstract":"Much data with graph structures satisfy the principle of homophily, meaning that connected nodes tend to be similar with respect to a specific attribute. As such, ubiquitous datasets for graph machine learning tasks have generally been highly homophilous, rewarding methods that leverage homophily as an inductive bias. Recent work has pointed out this particular focus, as new non-homophilous datasets have been introduced and graph representation learning models better suited for low-homophily settings have been developed. However, these datasets are small and poorly suited to truly testing the effectiveness of new methods in non-homophilous settings. We present a series of improved graph datasets with node label relationships that do not satisfy the homophily principle. Along with this, we introduce a new measure of the presence or absence of homophily that is better suited than existing measures in different regimes. We benchmark a range of simple methods and graph neural networks across our proposed datasets, drawing new insights for further research. Data and codes can be found at https://github.com/CUAI/Non-Homophily-Benchmarks.","url_abs":"https://arxiv.org/abs/2104.01404v2","url_pdf":"https://arxiv.org/pdf/2104.01404v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"new-benchmarks-for-learning-on-non","repo_url":"https://github.com/CUAI/Non-Homophily-Benchmarks","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"attribute","task_name":"Attribute"},{"task_slug":"fraud-detection","task_name":"Fraud Detection"},{"task_slug":"graph-representation-learning","task_name":"Graph Representation Learning"},{"task_slug":"inductive-bias","task_name":"Inductive Bias"},{"task_slug":"node-classification","task_name":"Node Classification"},{"task_slug":"node-classification-on-non-homophilic","task_name":"Node Classification on Non-Homophilic (Heterophilic) Graphs"},{"task_slug":"representation-learning","task_name":"Representation Learning"}],"methods":[],"datasets_introduced":[{"slug":"deezer-europe-1","name":"Deezer-Europe","full_name":""}],"methods_introduced":[],"results":[{"leaderboard":"/sota/fraud-detection-on-yelp-fraud","task":"Fraud Detection","dataset":"Yelp-Fraud","model":"GAT+JK","rank_in_archive_order":6,"of":10,"metrics":{"AUC-ROC":"90.04"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-penn94","task":"Node Classification","dataset":"Penn94","model":"GCNJK","rank_in_archive_order":18,"of":32,"metrics":{"Accuracy":"81.63 ± 0.54"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-penn94","task":"Node Classification","dataset":"Penn94","model":"LINK","rank_in_archive_order":22,"of":32,"metrics":{"Accuracy":"80.79 ± 0.49"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-penn94","task":"Node Classification","dataset":"Penn94","model":"GATJK","rank_in_archive_order":23,"of":32,"metrics":{"Accuracy":"80.69 ± 0.36"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-penn94","task":"Node Classification","dataset":"Penn94","model":"L Prop 2-hop","rank_in_archive_order":29,"of":32,"metrics":{"Accuracy":"74.13 ± 0.46"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-penn94","task":"Node Classification","dataset":"Penn94","model":"MLP","rank_in_archive_order":30,"of":32,"metrics":{"Accuracy":"73.61 ± 0.40"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-penn94","task":"Node Classification","dataset":"Penn94","model":"L Prop 1-hop","rank_in_archive_order":32,"of":32,"metrics":{"Accuracy":"63.21 ± 0.39"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-yelpchi","task":"Node Classification","dataset":"Yelp-Fraud","model":"GAT+JK","rank_in_archive_order":5,"of":9,"metrics":{"AUC-ROC":"90.04"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-genius","task":"Node Classification","dataset":"genius","model":"LINK","rank_in_archive_order":21,"of":26,"metrics":{"Accuracy":"73.56 ± 0.14"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-genius","task":"Node Classification","dataset":"genius","model":"L Prop 2-hop","rank_in_archive_order":22,"of":26,"metrics":{"Accuracy":"67.04 ± 0.20"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-genius","task":"Node Classification","dataset":"genius","model":"L Prop 1-hop","rank_in_archive_order":23,"of":26,"metrics":{"Accuracy":"66.02 ± 0.16"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-genius","task":"Node Classification","dataset":"genius","model":"GATJK","rank_in_archive_order":24,"of":26,"metrics":{"Accuracy":"56.70 ± 2.07"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-non-homophilic-6","task":"Node Classification on Non-Homophilic (Heterophilic) Graphs","dataset":"Deezer-Europe","model":"MLP-2","rank_in_archive_order":14,"of":28,"metrics":{"1:1 Accuracy":"66.55±0.72"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-non-homophilic-6","task":"Node Classification on Non-Homophilic (Heterophilic) Graphs","dataset":"Deezer-Europe","model":"GCN+JK","rank_in_archive_order":23,"of":28,"metrics":{"1:1 Accuracy":"60.99±0.14"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-non-homophilic-6","task":"Node Classification on Non-Homophilic (Heterophilic) Graphs","dataset":"Deezer-Europe","model":"GAT+JK","rank_in_archive_order":25,"of":28,"metrics":{"1:1 Accuracy":"59.66±0.92"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-non-homophilic-6","task":"Node Classification on Non-Homophilic (Heterophilic) Graphs","dataset":"Deezer-Europe","model":"LINK","rank_in_archive_order":26,"of":28,"metrics":{"1:1 Accuracy":"57.71±0.36"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-non-homophilic-6","task":"Node Classification on Non-Homophilic (Heterophilic) Graphs","dataset":"Deezer-Europe","model":"LProp (2hop)","rank_in_archive_order":27,"of":28,"metrics":{"1:1 Accuracy":"56.96±0.26"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-non-homophilic-6","task":"Node Classification on Non-Homophilic (Heterophilic) Graphs","dataset":"Deezer-Europe","model":"L Prop (1hop)","rank_in_archive_order":28,"of":28,"metrics":{"1:1 Accuracy":"56.50±0.41"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-non-homophilic-1","task":"Node Classification on Non-Homophilic (Heterophilic) Graphs","dataset":"Wisconsin(60%/20%/20% random splits)","model":"MLP-2","rank_in_archive_order":14,"of":32,"metrics":{"1:1 Accuracy":"93.87 ± 3.33"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2104.01404","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}