{"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/relation-prediction-as-an-auxiliary-training","title":"Relation Prediction as an Auxiliary Training Objective for Improving Multi-Relational Graph Representations","arxiv_id":"2110.02834","date":"2021-10-06","proceeding":"AKBC 2021 10","authors":["Yihong Chen","Pasquale Minervini","Sebastian Riedel","Pontus Stenetorp"],"abstract":"Learning good representations on multi-relational graphs is essential to knowledge base completion (KBC). In this paper, we propose a new self-supervised training objective for multi-relational graph representation learning, via simply incorporating relation prediction into the commonly used 1vsAll objective. The new training objective contains not only terms for predicting the subject and object of a given triple, but also a term for predicting the relation type. We analyse how this new objective impacts multi-relational learning in KBC: experiments on a variety of datasets and models show that relation prediction can significantly improve entity ranking, the most widely used evaluation task for KBC, yielding a 6.1% increase in MRR and 9.9% increase in Hits@1 on FB15k-237 as well as a 3.1% increase in MRR and 3.4% in Hits@1 on Aristo-v4. Moreover, we observe that the proposed objective is especially effective on highly multi-relational datasets, i.e. datasets with a large number of predicates, and generates better representations when larger embedding sizes are used.","url_abs":"https://arxiv.org/abs/2110.02834v1","url_pdf":"https://arxiv.org/pdf/2110.02834v1.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":"relation-prediction-as-an-auxiliary-training","repo_url":"https://github.com/facebookresearch/ssl-relation-prediction","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"NOASSERTION"}}],"tasks":[{"task_slug":"graph-representation-learning","task_name":"Graph Representation Learning"},{"task_slug":"knowledge-base-completion","task_name":"Knowledge Base Completion"},{"task_slug":"knowledge-graph-completion","task_name":"Knowledge Graph Completion"},{"task_slug":"knowledge-graph-embedding","task_name":"Knowledge Graph Embedding"},{"task_slug":"knowledge-graph-embeddings","task_name":"Knowledge Graph Embeddings"},{"task_slug":"link-prediction","task_name":"Link Prediction"},{"task_slug":"link-property-prediction","task_name":"Link Property Prediction"},{"task_slug":null,"task_name":"Relation"},{"task_slug":"relation-prediction","task_name":"Relation Prediction"},{"task_slug":"relational-reasoning","task_name":"Relational Reasoning"}],"methods":[{"method_slug":"cp-n3","method_name":"CP-N3"},{"method_slug":"cp-n3-rp-1","method_name":"CP-N3-RP"},{"method_slug":"complex-n3","method_name":"ComplEx-N3"},{"method_slug":"complex-n3-rp","method_name":"ComplEx-N3-RP"},{"method_slug":"rescal-rp","method_name":"RESCAL"},{"method_slug":"rescal-rp-1","method_name":"RESCAL-RP"},{"method_slug":"tucker","method_name":"TuckER"},{"method_slug":"tucker-rp","method_name":"TuckER-RP"}],"datasets_introduced":[],"methods_introduced":[{"slug":"cp-n3-rp-1","name":"CP-N3-RP","full_name":"CP with N3 Regularizer and Relation Prediction"},{"slug":"complex-n3-rp","name":"ComplEx-N3-RP","full_name":"ComplEx with N3 Regularizer and Relation Prediction Objective"},{"slug":"rescal-rp-1","name":"RESCAL-RP","full_name":"RESCAL with Relation Prediction"},{"slug":"tucker-rp","name":"TuckER-RP","full_name":"TuckER with Relation Prediction"}],"results":[{"leaderboard":"/sota/link-prediction-on-aristo-v4","task":"Link Prediction","dataset":"Aristo-v4","model":"ComplEx-N3-RP","rank_in_archive_order":1,"of":1,"metrics":{"Hits@1":"0.24","Hits@10":"0.447","Hits@3":"0.336","MRR":"0.311"},"uses_additional_data":false},{"leaderboard":"/sota/link-prediction-on-codex-large","task":"Link Prediction","dataset":"CoDEx Large","model":"ComplEx-N3-RP","rank_in_archive_order":1,"of":6,"metrics":{"Hits@1":"0.277","Hits@10":"0.473","Hits@3":"0.377","MRR":"0.345"},"uses_additional_data":false},{"leaderboard":"/sota/link-prediction-on-codex-medium","task":"Link Prediction","dataset":"CoDEx Medium","model":"ComplEx-N3-RP","rank_in_archive_order":2,"of":7,"metrics":{"Hits@1":"0.277","Hits@10":"0.490","Hits@3":"0.386","MRR":"0.352"},"uses_additional_data":false},{"leaderboard":"/sota/link-prediction-on-codex","task":"Link Prediction","dataset":"CoDEx Small","model":"ComplEx-N3-RP","rank_in_archive_order":1,"of":6,"metrics":{"Hits@1":"0.375","Hits@10":"0.663","Hits@3":"0.514","MRR":"0.473"},"uses_additional_data":false},{"leaderboard":"/sota/link-prediction-on-fb15k-237","task":"Link Prediction","dataset":"FB15k-237","model":"ComplEx-N3-RP","rank_in_archive_order":5,"of":75,"metrics":{"Hits@1":"0.298","Hits@10":"0.568","Hits@3":"0.424","MR":"163","MRR":"0.389"},"uses_additional_data":false},{"leaderboard":"/sota/link-prediction-on-fb15k-237","task":"Link Prediction","dataset":"FB15k-237","model":"TuckER-RP","rank_in_archive_order":26,"of":75,"metrics":{"Hits@1":"0.264","Hits@10":"0.535","Hits@3":"0.388","MRR":"0.354"},"uses_additional_data":false},{"leaderboard":"/sota/link-prediction-on-fb15k-237","task":"Link Prediction","dataset":"FB15k-237","model":"CP-N3-RP","rank_in_archive_order":59,"of":75,"metrics":{"Hits@10":"0.55","MRR":"0.366"},"uses_additional_data":false},{"leaderboard":"/sota/link-prediction-on-wn18rr","task":"Link Prediction","dataset":"WN18RR","model":"ComplEx-N3-RP","rank_in_archive_order":29,"of":75,"metrics":{"Hits@1":"0.443","Hits@10":"0.578","Hits@3":"0.505","MRR":"0.488"},"uses_additional_data":false},{"leaderboard":"/sota/link-property-prediction-on-ogbl-biokg","task":"Link Property Prediction","dataset":"ogbl-biokg","model":"ComplEx-N3-RP","rank_in_archive_order":7,"of":16,"metrics":{"Number of params":"187750000","Test MRR":"0.8494","Validation MRR":"0.8497"},"uses_additional_data":false},{"leaderboard":"/sota/link-property-prediction-on-ogbl-wikikg2","task":"Link Property Prediction","dataset":"ogbl-wikikg2","model":"ComplEx-N3-RP (100dim)","rank_in_archive_order":13,"of":30,"metrics":{"Number of params":"500334800","Test MRR":"0.6481","Validation MRR":"0.6701"},"uses_additional_data":false},{"leaderboard":"/sota/link-property-prediction-on-ogbl-wikikg2","task":"Link Property Prediction","dataset":"ogbl-wikikg2","model":"ComplEx-N3-RP (50dim)","rank_in_archive_order":15,"of":30,"metrics":{"Number of params":"250167400","Test MRR":"0.6364","Validation MRR":"0.6594"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2110.02834","atlas_url":"https://app.syntology.ai/?focus=2110.02834","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}