{"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/neural-ranking-models-with-weak-supervision","title":"Neural Ranking Models with Weak Supervision","arxiv_id":"1704.08803","date":"2017-04-28","proceeding":null,"authors":["Mostafa Dehghani","Hamed Zamani","Aliaksei Severyn","Jaap Kamps","W. Bruce Croft"],"abstract":"Despite the impressive improvements achieved by unsupervised deep neural\nnetworks in computer vision and NLP tasks, such improvements have not yet been\nobserved in ranking for information retrieval. The reason may be the complexity\nof the ranking problem, as it is not obvious how to learn from queries and\ndocuments when no supervised signal is available. Hence, in this paper, we\npropose to train a neural ranking model using weak supervision, where labels\nare obtained automatically without human annotators or any external resources\n(e.g., click data). To this aim, we use the output of an unsupervised ranking\nmodel, such as BM25, as a weak supervision signal. We further train a set of\nsimple yet effective ranking models based on feed-forward neural networks. We\nstudy their effectiveness under various learning scenarios (point-wise and\npair-wise models) and using different input representations (i.e., from\nencoding query-document pairs into dense/sparse vectors to using word embedding\nrepresentation). We train our networks using tens of millions of training\ninstances and evaluate it on two standard collections: a homogeneous news\ncollection(Robust) and a heterogeneous large-scale web collection (ClueWeb).\nOur experiments indicate that employing proper objective functions and letting\nthe networks to learn the input representation based on weakly supervised data\nleads to impressive performance, with over 13% and 35% MAP improvements over\nthe BM25 model on the Robust and the ClueWeb collections. Our findings also\nsuggest that supervised neural ranking models can greatly benefit from\npre-training on large amounts of weakly labeled data that can be easily\nobtained from unsupervised IR models.","url_abs":"http://arxiv.org/abs/1704.08803v2","url_pdf":"http://arxiv.org/pdf/1704.08803v2.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":"neural-ranking-models-with-weak-supervision","repo_url":"https://github.com/mikvrax/TrecingLab","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"ad-hoc-information-retrieval","task_name":"Ad-Hoc Information Retrieval"},{"task_slug":"information-retrieval","task_name":"Information Retrieval"},{"task_slug":"retrieval","task_name":"Retrieval"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/ad-hoc-information-retrieval-on-trec-robust04","task":"Ad-Hoc Information Retrieval","dataset":"TREC Robust04","model":"FNRM-RankProb_Embed","rank_in_archive_order":19,"of":21,"metrics":{"MAP":"0.2837"},"uses_additional_data":false},{"leaderboard":"/sota/ad-hoc-information-retrieval-on-trec-robust04","task":"Ad-Hoc Information Retrieval","dataset":"TREC Robust04","model":"FNRM-Rank_Embed","rank_in_archive_order":20,"of":21,"metrics":{"MAP":"0.2811"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1704.08803","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}