[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85811-en":3,"doc-seo-85811-105":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},85811,8796095461564,"Liam","https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d",8,"Research & Report","Scaling and Stabilizing Large-Scale Embedding-Based Retrieval","Embedding-based retrieval (EBR) is essential for large-scale ecommerce search but is often limited by training signal quality and encoder capacity. Dual-encoder training faces a training-inference gap, since models are optimized on narrow candidate sets yet must separate hundreds of millions of items at inference. Upgrading backbones alone can also destabilize behavior and erase accumulated domain knowledge. A unified Walmart pipeline combines hybrid hard negative mining and legacy-aware distillation to improve both relevance and robustness, delivering measured gains in ranking and revenue.","Scaling and Stabilizing Large-Scale Embedding-Based Retrieval  \nZhen Yang∗† Walmart Global Tech Sunnyvale, USA[njuyangz@gmail.com](njuyangz@gmail.com)  \nFeng Liu∗ Walmart Global Tech Sunnyvale, USA [liu.f@pku.edu.cn](liu.f@pku.edu.cn)  \nJuexin Lin∗† Walmart Global Tech Sunnyvale, USA [linjuexin@gmail.com](linjuexin@gmail.com)  \nSatya Chembolu  \nWalmart Global Tech Sunnyvale, USA  \nHongwei Shang  \nWalmart Global Tech Sunnyvale, USA [hongwei.shang@walmart.com](hongwei.shang@walmart.com)  \nXunfan Cai  \nWalmart Global Tech Sunnyvale, USA  \n[satya.chembolu@walmart.com](satya.chembolu@walmart.com) [xunfan.cai@walmart.com](xunfan.cai@walmart.com)  \nKaihao Li  \nWalmart Global Tech Sunnyvale, USA [kaihao.li@walmart.com](kaihao.li@walmart.com)  \nXinyi Liu  \nWalmart Global Tech Sunnyvale, USA [xinyi.liu@walmart.com](xinyi.liu@walmart.com)  \nCun Mu  \nWalmart Global Tech Sunnyvale, USA[cun.mu@walmart.com](cun.mu@walmart.com)  \nTony Lee∗ Walmart Global Tech Sunnyvale, USA [tonyleexyz@gmail.com](tonyleexyz@gmail.com)  \nCiya Liao∗ Walmart Global Tech Sunnyvale, USA [liao_ciya@yahoo.com](liao_ciya@yahoo.com)  \narXiv :2607 . 10096v 1 [ cs .IR] 11 Jul 2026  \nAbstract  \nEmbedding-based retrieval (EBR) is foundational to large-scale ecommerce search, yet its effectiveness is often constrained by the quality of training signals and the representational capacity of the encoder. Standard dual-encoders suffer from a training-inference gap: they are optimized on narrow candidate pools but must discriminate against hundreds of millions of items during inference. Furthermore, while transitioning to higher-capacity backbones can mitigate this gap, simply replacing a mature model can lead to inconsistent retrieval behavior and a loss of the domain-specific knowledge established in previous iterations. In this paper, we present a unified pipeline deployed at Walmart that addresses both signal quality and model evolution. Our contributions are two-fold:  \n(1) Hybrid Hard Negative Mining: We integrate Online Cross-Batch Sampling to increase negative diversity by an order of magnitude and Hybrid Offline Mining, which combines cross-encoder predictions with metadata heuristics to identify nuanced mismatches.  \n(2) Legacy-Aware Distillation: We transition from DistilBERT to a higher-capacity GTE-base encoder. To ensure a smooth and superior transition, we introduce a Warm-Start Distillation technique that transfers domain-specific expertise from the legacy model to the new backbone. Validated through extensive offline experiments and online A/B testing, the proposed pipeline is deployed in live production, delivering a +7.34% improvement in NDCG@5 and a +0.50% lift in gross revenue.  \nCCS Concepts  \n• Information systems → Retrieval models and ranking.  \n∗ Work done while at Walmart.  \n†Both authors contributed equally to this research.  \nThis work is licensed under a Creative Commons Attribution 4 .0 International License. SIGIR ’26, Melbourne, VIC, Australia  \n© 2026 Copyright held by the owner/author(s) .  \nACM ISBN 979-8-4007-2599-9/2026/07  \n[https://doi.org/10.1145/3805712.3808474](https://doi.org/10.1145/3805712.3808474)  \nKeywords  \nEmbedding-based Retrieval, Hard Negative Mining, Knowledge Distillation, E-commerce Search, Dense Retrieval, Dual-Encoders  \nACM Reference Format:  \nZhen Yang, Juexin Lin, Hongwei Shang, Kaihao Li, Feng Liu, Satya Chembolu, Xunfan Cai, Xinyi Liu, Cun Mu, Tony Lee, and Ciya Liao. 2026. Scaling and Stabilizing Large-Scale Embedding-Based Retrieval. In Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’26), July 20–24, 2026, Melbourne, VIC, Australia. ACM, New York, NY, USA, 6 pages. [https://doi.org/10.1145/3805712.3808474](https://doi.org/10.1145/3805712.3808474)  \n1 Introduction  \nEmbedding-based retrieval (EBR) using dual-encoder architecturesand Approximate Nearest Neighbor (ANN) [10] search has become a foundational component of large-scale e-commerce pla","cbCaiukzmJYzYYLj","https://ap.wps.com/l/cbCaiukzmJYzYYLj","pdf",822963,1,6,"English","en",105,"# Abstract\n# Keywords\n# Introduction","[{\"question\":\"What problem does the paper target in large-scale embedding-based retrieval?\",\"answer\":\"It targets the training-inference gap and representation-capacity limits, where dual-encoder models trained on limited candidate pools must discriminate among extremely large catalogs during inference.\"},{\"question\":\"How does the proposed approach improve hard negative diversity?\",\"answer\":\"It combines Online Cross-Batch Sampling to expand negative diversity across GPUs with Hybrid Offline Mining that uses cross-encoder predictions plus metadata heuristics to find subtle mismatches.\"},{\"question\":\"How does the paper handle model upgrades without losing domain knowledge?\",\"answer\":\"It introduces Legacy-Aware Distillation, using Warm-Start Distillation to transfer domain-specific expertise from a legacy DistilBERT model to a higher-capacity GTE-base encoder, validated with offline experiments and online A/B testing.\"}]",1784206406,15,{"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},"scaling-and-stabilizing-large-scale-embedding-based-retrieval","",{"@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/scaling-and-stabilizing-large-scale-embedding-based-retrieval/85811/",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 the paper target in large-scale embedding-based retrieval?","Question",{"text":75,"@type":76},"It targets the training-inference gap and representation-capacity limits, where dual-encoder models trained on limited candidate pools must discriminate among extremely large catalogs during inference.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the proposed approach improve hard negative diversity?",{"text":80,"@type":76},"It combines Online Cross-Batch Sampling to expand negative diversity across GPUs with Hybrid Offline Mining that uses cross-encoder predictions plus metadata heuristics to find subtle mismatches.",{"name":82,"@type":73,"acceptedAnswer":83},"How does the paper handle model upgrades without losing domain knowledge?",{"text":84,"@type":76},"It introduces Legacy-Aware Distillation, using Warm-Start Distillation to transfer domain-specific expertise from a legacy DistilBERT model to a higher-capacity GTE-base encoder, validated with offline experiments and online A/B testing.","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},[93,97,101,105,110,114,119,122,127,130,134],{"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":21,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},"Technology",50,"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":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]