[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82750-en":3,"doc-seo-82750-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},82750,549758252649,"Ivy","https://ap-avatar.wpscdn.com/avatar/8000253669c5317157?_k=1778319167496531819",8,"Research & Report","Relevance-Based Embeddings Lightweight Candidate Retrieval via Heavy-Ranker Calls","Relevance-Based Embeddings presents a lightweight candidate retrieval approach for machine learning search pipelines where computing exact relevance with an expensive similarity model is too costly for exhaustive comparison. Instead of embedding queries and items directly, the method derives new representations from the expensive model’s relevance scores to a selected support set. Theoretically, the resulting embeddings approximate arbitrary complex similarity functions under mild conditions. Experiments on academic and production datasets demonstrate improved candidate selection via carefully designed support selection strategies.","Relevance-Based Embeddings: Lightweight Candidate Retrieval via Heavy-Ranker Calls  \nKirill Shevkunov 1 Andrey Ploskonosov 1 Liudmila Prokhorenkova 1  \narXiv :2607 .035 15v 1 [ cs .IR] 3 Jul 2026  \nAbstract  \nIn many machine learning applications, the most relevant items for a query should be efficiently retrieved. The relevance function is usually an expensive similarity model, making the exhaustive search infeasible. A typical solution is to train another model that separately embeds queries and items to a vector space, where similarity is defined via the dot product or cosine similarity. This allows one to search the relevant items through fast approximate nearest neighbor search at the cost of some reduction in quality. To compensate for this reduction, the found items (candidates) are re-ranked by the expensive ranking model. In this paper, we investigate an alternative approach to candidate selection that utilizes the scores of the expensive model to improve the representations of queries and items. The idea is to describe each query (item) by its relevance to a set of support items (queries) and use these new representations to obtain query (item) embeddings. We theoretically prove that such embeddings are powerful enough to approximate any complex similarity model (under mild conditions) . We also investigate the choice of support items, which is a crucial ingredient of the proposed approach. The experiments on diverse academic and production datasets illustrate the power of our method.  \n1. Introduction  \nFinding the most relevant element (item) i to a query q among a large set of candidates I is a key task for a wide range of machine learning problems, for example, information retrieval, recommender systems, question-answering systems, or search engines. In such problems, the final  \n1Yandex. Correspondence to: Kirill Shevkunov \u003C[shevkunov@yandex-team.ru](shevkunov@yandex-team.ru) >, Liudmila Prokhorenkova  \n\u003C[ostroumova-la@yandex-team.ru](ostroumova-la@yandex-team.ru) >.  \nProceedings of the 43 rd International Conference on Machine Learning, Seoul, South Korea. PMLR 306, 2026 . Copyright 2026 by the author(s) .  \nscore (relevance) is often predicted by a pairwise function R : I × Q → R, where Q is a query space and R approximates some ground truth relevance such as click probability or time spent. Depending on the task, the relevance function R can utilize query attributes (e.g., the text of the query or a set of numerical features describing the user, such as age, time spent on the service, etc.), item attributes, or attributes describing the query-item pair (e.g., statistics based on counts of each query term in the document in information retrieval tasks) .  \nThe problem of retrieving the most relevant item for a query q can be written as arg maxi∈I R (i, q) . For practical applications, it is usually required to return not one but K best items (for directly displaying to the user or further re-ranking) . Most recommender systems have large item spaces I (millions to hundreds of millions), so exhaustive search is infeasible. Thi˜s problem is often solved  \nby training an auxiliary model R, called a Siamese, twotower, ˜or dual encoder (DE), in which late binding is  \nused: R (i, q) = S (FI (i), FQ (q)), where FI : I → Rd , FQ : Q → Rd , and S is some lightweight similarity measure, usually dot product or cosine similarity.  \nWhile a lot of effort has been put into developing dualencoder models, the cross-encoder (CE) ones are generally more powerful (Wu et al., 2020 ; Yadav et al., 2022) . Moreover, in practice it is typical to also have features that describe a query-item pair: e.g., counts of query terms in the document (information retrieval), information about previous user-item interactions (recommender systems), and soon. Such features cannot be used by dual encoders thus limiting their expressivity.  \nAn alternative approach suggested by Yadav et al. (2022) is to approximate the relevance of a given query ","cbCaibiuHlXotkcE","https://ap.wps.com/l/cbCaibiuHlXotkcE","pdf",2129773,1,20,"English","en",105,"# Abstract\n# Introduction\n## Problem setting: relevance retrieval and top-K search\n## Dual encoders versus cross encoders\n## Relevance-based embeddings idea\n## Choosing support items\n## Theoretical analysis","[{\"question\":\"What problem does Relevance-Based Embeddings address?\",\"answer\":\"It targets efficient retrieval of the most relevant items for a query when the true relevance model is expensive, making exhaustive search infeasible.\"},{\"question\":\"How does the method build query and item embeddings?\",\"answer\":\"It describes each query (item) through its relevance to a pre-selected set of support items (support queries) using scores from the expensive model, then uses these new representations for retrieval.\"},{\"question\":\"What is the role of selecting support items?\",\"answer\":\"Support selection is identified as a crucial component; the paper compares simple heuristics and more advanced strategies, showing that clustering-based or optimized choices can significantly improve relevance approximation quality.\"}]",1784182679,50,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"relevance-based-embeddings-lightweight-candidate-retrieval-via-heavy-ranker-calls","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/relevance-based-embeddings-lightweight-candidate-retrieval-via-heavy-ranker-calls/82750/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does Relevance-Based Embeddings address?","Question",{"text":75,"@type":76},"It targets efficient retrieval of the most relevant items for a query when the true relevance model is expensive, making exhaustive search infeasible.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the method build query and item embeddings?",{"text":80,"@type":76},"It describes each query (item) through its relevance to a pre-selected set of support items (support queries) using scores from the expensive model, then uses these new representations for retrieval.",{"name":82,"@type":73,"acceptedAnswer":83},"What is the role of selecting support items?",{"text":84,"@type":76},"Support selection is identified as a crucial component; the paper compares simple heuristics and more advanced strategies, showing that clustering-based or optimized choices can significantly improve relevance approximation quality.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,114,119,122,126,129,133],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":28,"slug":113},6,"Technology","technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":21,"slug":125},9,"Religion & Spirituality","religion-spirituality",{"id":21,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":21,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":106,"slug":136},19,"General","general"]