Proud to share our work, CoGR (Co-evolving Generative Retrieval), where we train LLMs to generate retrieval keywords for both user queries and apps.</p>\n<p>CoGR follows a co-evolving training framework: we alternately optimize one side while keeping the other side fixed, allowing the query and app keywords to gradually adapt to each other. Through this iterative process, we observe steady improvements in retrieval performance.</p>\n","updatedAt":"2026-09-03T01:33:11.063Z","author":{"_id":"65037565da2d88e201f63b7a","avatarUrl":"/avatars/d1b6ce17236360e9583b8bb4cb87e506.svg","fullname":"Runpeng Dai","name":"Leo-Dai","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8770807981491089},"editors":["Leo-Dai"],"editorAvatarUrls":["/avatars/d1b6ce17236360e9583b8bb4cb87e506.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2609.00638","authors":[{"_id":"6a98ce24fea818274321fdf1","user":{"_id":"65037565da2d88e201f63b7a","avatarUrl":"/avatars/d1b6ce17236360e9583b8bb4cb87e506.svg","isPro":false,"fullname":"Runpeng Dai","user":"Leo-Dai","type":"user","name":"Leo-Dai"},"name":"Runpeng Dai","status":"claimed_verified","statusLastChangedAt":"2026-09-03T08:17:15.886Z","hidden":false},{"_id":"6a98ce24fea818274321fdf2","user":{"_id":"62c61f82ad85f7873afc15b6","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62c61f82ad85f7873afc15b6/p6QwiGwK1AlvMo4Nb4GCN.jpeg","isPro":false,"fullname":"Kaili Huang","user":"khuang","type":"user","name":"khuang"},"name":"Kaili Huang","status":"claimed_verified","statusLastChangedAt":"2026-09-03T08:17:14.006Z","hidden":false},{"_id":"6a98ce24fea818274321fdf3","name":"Changsung Kang","hidden":false},{"_id":"6a98ce24fea818274321fdf4","name":"Ciya Liao","hidden":false}],"publishedAt":"2026-09-01T00:00:00.000Z","submittedOnDailyAt":"2026-09-03T00:00:00.000Z","title":"It Takes Two to Match: Co-Evolving Generative Retriever with Reinforcement Learning","submittedOnDailyBy":{"_id":"65037565da2d88e201f63b7a","avatarUrl":"/avatars/d1b6ce17236360e9583b8bb4cb87e506.svg","isPro":false,"fullname":"Runpeng Dai","user":"Leo-Dai","type":"user","name":"Leo-Dai"},"summary":"Retrieval is the first stage of modern search and advertising systems, selecting a candidate set from a large item universe for downstream ranking and auction. Recent work increasingly leverages LLMs to improve retrieval through query expansion, data synthesis, and retrieval-feedback training. However, the generative component is typically used for query-side augmentation, while final matching is still delegated to a downstream retriever. We introduce CoGR, a retrieval framework that instead trains LLMs to directly construct retrieval representations on both query and item sides. Each generator produces a compact set of keywords, which are matched directly through an inverted index, preserving compatibility with existing keyword-based retrieval infrastructure. CoGR uses a two-stage training pipeline. Supervised fine-tuning first establishes an aligned keyword space, after which co-evolving reinforcement learning alternately optimizes the query- and item-side generators with GRPO against the opposite side's frozen index. Both sides optimize the same query-to-item retrieval F_1 objective: the query side receives retrieval F_1 directly, while the item side receives a counterfactual marginal reward measuring the change in query-side F_1 caused by its generated keywords. Across 10 representative sparse, dense, and generative baselines, CoGR achieves the best performance on both an internal APP Marketplace dataset and the public WANDS benchmark, improving F_1 over the strongest baseline by 10.9% and 36.1%, respectively. Further analysis shows stable co-evolution and increasingly aligned query--item keyword spaces over training.","upvotes":60,"discussionId":"6a98ce24fea818274321fdf5","ai_summary":"CoGR trains LLMs to generate compact keywords for both queries and items, enabling direct inverted-index retrieval optimized via co-evolving reinforcement learning.","ai_keywords":["LLMs","query expansion","retrieval-feedback training","CoGR","retrieval representations","inverted index","supervised fine-tuning","co-evolving reinforcement learning","GRPO","counterfactual marginal reward","sparse retrieval","dense retrieval","generative retrieval"],"ai_summary_model":"thinkingmachines/Inkling-Small","organization":{"_id":"628cbd99ef14f971b69948ab","name":"apple","fullname":"Apple","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1653390727490-5dd96eb166059660ed1ee413.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"65037565da2d88e201f63b7a","avatarUrl":"/avatars/d1b6ce17236360e9583b8bb4cb87e506.svg","isPro":false,"fullname":"Runpeng Dai","user":"Leo-Dai","type":"user"},{"_id":"6a7f0dce08eba9f843df1e19","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a7f0dce08eba9f843df1e19/zDODRgsnbsg6ziKqdfy1N.jpeg","isPro":false,"fullname":"Elijah F. Jones","user":"Eljones2007","type":"user"},{"_id":"6623ea65b642e29cdf90a1b4","avatarUrl":"/avatars/e32e90574c1162b2be87ed78604e3e4d.svg","isPro":false,"fullname":"TongZheng","user":"TongZheng1999","type":"user"},{"_id":"6a7dd17f6610798c43264809","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a7dd17f6610798c43264809/be5T1lUcU_kwW2Uu6T1-Q.jpeg","isPro":false,"fullname":"Lucía Oliveira-Costa","user":"vikasreddyson","type":"user"},{"_id":"6a824e391b36723daa5834ea","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a824e391b36723daa5834ea/jZiVYtaiZrioF8tPkbDIX.jpeg","isPro":false,"fullname":"प्रिया पटेल","user":"Aaravpat","type":"user"},{"_id":"62ea79dd01ed9b0e8f61ccd3","avatarUrl":"/avatars/70af83e0e267be39fcd5f23b85e2dafa.svg","isPro":false,"fullname":"Chengsong Huang","user":"ChengsongHuang","type":"user"},{"_id":"6a81454a818008a8c96ee7ef","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a81454a818008a8c96ee7ef/atVuTlBEYqbadb_cWXs8l.jpeg","isPro":false,"fullname":"吴欣怡","user":"austin-hall","type":"user"},{"_id":"6a833255388a811354380df8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a833255388a811354380df8/JfjoutlWxnWrWsqkaP46S.jpeg","isPro":false,"fullname":"Xie Ivan","user":"itsiv-anxie","type":"user"},{"_id":"64e6c617ecce34cb442cb208","avatarUrl":"/avatars/ebc61bf6a043314cb2089b1efd5e6a18.svg","isPro":false,"fullname":"JieSun(SII)","user":"Sunshine279","type":"user"},{"_id":"6850ab3fb73a72cdc6159f6f","avatarUrl":"/avatars/74c58336655cdfe7d3fb9854d426240d.svg","isPro":false,"fullname":"Haolin Liu","user":"lhl616","type":"user"},{"_id":"62c61f82ad85f7873afc15b6","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62c61f82ad85f7873afc15b6/p6QwiGwK1AlvMo4Nb4GCN.jpeg","isPro":false,"fullname":"Kaili Huang","user":"khuang","type":"user"},{"_id":"6a882a7ec5707b343fca644e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a882a7ec5707b343fca644e/U9DRHYIlNQ7ld_SshsmWa.jpeg","isPro":false,"fullname":"Mohammed Nasser","user":"rydn-asser","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"628cbd99ef14f971b69948ab","name":"apple","fullname":"Apple","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1653390727490-5dd96eb166059660ed1ee413.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2609/2609.00638.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2609.00638 in a model README.md to link it from this page.

Cite arxiv.org/abs/2609.00638 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2609.00638 in a Space README.md to link it from this page.