{"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/learning-edge-representations-via-low-rank","title":"Learning Edge Representations via Low-Rank Asymmetric Projections","arxiv_id":"1705.05615","date":"2017-05-16","proceeding":null,"authors":["Sami Abu-El-Haija","Bryan Perozzi","Rami Al-Rfou"],"abstract":"We propose a new method for embedding graphs while preserving directed edge\ninformation. Learning such continuous-space vector representations (or\nembeddings) of nodes in a graph is an important first step for using network\ninformation (from social networks, user-item graphs, knowledge bases, etc.) in\nmany machine learning tasks.\n  Unlike previous work, we (1) explicitly model an edge as a function of node\nembeddings, and we (2) propose a novel objective, the \"graph likelihood\", which\ncontrasts information from sampled random walks with non-existent edges.\nIndividually, both of these contributions improve the learned representations,\nespecially when there are memory constraints on the total size of the\nembeddings. When combined, our contributions enable us to significantly improve\nthe state-of-the-art by learning more concise representations that better\npreserve the graph structure.\n  We evaluate our method on a variety of link-prediction task including social\nnetworks, collaboration networks, and protein interactions, showing that our\nproposed method learn representations with error reductions of up to 76% and\n55%, on directed and undirected graphs. In addition, we show that the\nrepresentations learned by our method are quite space efficient, producing\nembeddings which have higher structure-preserving accuracy but are 10 times\nsmaller.","url_abs":"http://arxiv.org/abs/1705.05615v4","url_pdf":"http://arxiv.org/pdf/1705.05615v4.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":"learning-edge-representations-via-low-rank","repo_url":"https://github.com/google/asymproj_edge_dnn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"link-prediction","task_name":"Link Prediction"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1705.05615","atlas_url":"https://app.syntology.ai/?focus=1705.05615","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}