RecHarness is a bandit-routed agentic harness that splits recommender-model optimization into \"bandit picks the direction, LLM generates the hypothesis and code,\" achieving more stable, budget-efficient gains than pure LLM-reasoning search.</p>\n","updatedAt":"2026-08-04T09:08:55.764Z","author":{"_id":"6683b38720ee9ac417195830","avatarUrl":"/avatars/6066e5dfb8cf7e54de18f013fe4d1da6.svg","fullname":"Yuecheng Li","name":"6lyc","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9033330678939819},"editors":["6lyc"],"editorAvatarUrls":["/avatars/6066e5dfb8cf7e54de18f013fe4d1da6.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2607.29241","authors":[{"_id":"6a71abc1205150e09f8b145a","name":"Haoran Ling","hidden":false},{"_id":"6a71abc1205150e09f8b145b","user":{"_id":"6683b38720ee9ac417195830","avatarUrl":"/avatars/6066e5dfb8cf7e54de18f013fe4d1da6.svg","isPro":false,"fullname":"Yuecheng Li","user":"6lyc","type":"user","name":"6lyc"},"name":"Yuecheng Li","status":"claimed_verified","statusLastChangedAt":"2026-08-04T10:00:41.160Z","hidden":false},{"_id":"6a71abc1205150e09f8b145c","name":"Zeyu Song","hidden":false},{"_id":"6a71abc1205150e09f8b145d","name":"Jing Yao","hidden":false},{"_id":"6a71abc1205150e09f8b145e","name":"Shuwen Kang","hidden":false},{"_id":"6a71abc1205150e09f8b145f","name":"Chi Lu","hidden":false},{"_id":"6a71abc1205150e09f8b1460","name":"Wenjin Wu","hidden":false},{"_id":"6a71abc1205150e09f8b1461","name":"Peng Jiang","hidden":false}],"publishedAt":"2026-07-31T00:00:00.000Z","submittedOnDailyAt":"2026-08-04T00:00:00.000Z","title":"RecHarness: A Bandit-Routed Agentic Harness for Self-Evolving Recommender Systems","submittedOnDailyBy":{"_id":"6683b38720ee9ac417195830","avatarUrl":"/avatars/6066e5dfb8cf7e54de18f013fe4d1da6.svg","isPro":false,"fullname":"Yuecheng Li","user":"6lyc","type":"user","name":"6lyc"},"summary":"Optimizing modern recommender models still depends heavily on engineers manually iterating over architectural, objective, and training-strategy changes. While LLM-based agents can automate this trial-and-error process, allowing the LLM to both select modification directions and generate concrete hypotheses often leads to unstable search under limited experiment budgets. Inspired by the above challenge, we propose RecHarness, a Bandit-Routed Agentic Harness for automated recommender model optimization. RecHarness separates the optimization process into two steps: a bandit router selects the next modification direction according to historical validation feedback, while the LLM generates a concrete optimization hypothesis and executable code edit within the selected direction. To sustain long-horizon exploration, RecHarness uses a jump-basin mechanism to activate a structural-jump arm when local edits stagnate. Across multiple recommendation tasks, datasets, and model backbones, RecHarness achieves more stable performance improvements and uses limited trial budgets more effectively than LLM-reasoning search. During a 7-day online A/B test on a large-scale short-video advertising platform, the selected candidate improves ADVV by 2.084%, Revenue by 0.534%, and Exposure by 0.559%. Code is available at https://github.com/6lyc/RecHarness.","upvotes":6,"discussionId":"6a71abc1205150e09f8b1462","githubRepo":"https://github.com/6lyc/RecHarness","githubRepoAddedBy":"user","githubStars":1,"organization":{"_id":"6611fef5cfbba5e7612b96b5","name":"kuaishou","fullname":"kuaishou technology","avatar":"https://www.gravatar.com/avatar/e54df970bc4d3c2ab62cf1dfa4b78559?d=retro&size=100"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6683b38720ee9ac417195830","avatarUrl":"/avatars/6066e5dfb8cf7e54de18f013fe4d1da6.svg","isPro":false,"fullname":"Yuecheng Li","user":"6lyc","type":"user"},{"_id":"6a6c7a3d3574d63d5305cd74","avatarUrl":"/avatars/78a0b7bd8c1b9996c19966b58df44488.svg","isPro":false,"fullname":"Richard Wilson","user":"Cobalt-Richard4","type":"user"},{"_id":"6a6c7b702b8f6bcb1b61caba","avatarUrl":"/avatars/40259c5b4eb7b81d54b23737398a112f.svg","isPro":false,"fullname":"Sarah Clark","user":"sarah-clark","type":"user"},{"_id":"6a6c7a12a7b7e4e7d009ab72","avatarUrl":"/avatars/56fe1f3ce4693a21ac34744b48675e17.svg","isPro":false,"fullname":"Sarah Martin","user":"benjamin-4988438","type":"user"},{"_id":"6a6da95b346df769d90cc578","avatarUrl":"/avatars/d8fb17b7c3265f6df50052a620fa774f.svg","isPro":false,"fullname":"Daniel Clark","user":"john-8480530","type":"user"},{"_id":"69bb59f0aae0d99fe476a9a5","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/Sh-5dqK0wSv4NMKvL2TJN.png","isPro":false,"fullname":"James Johnson","user":"wyattbaker","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"6611fef5cfbba5e7612b96b5","name":"kuaishou","fullname":"kuaishou technology","avatar":"https://www.gravatar.com/avatar/e54df970bc4d3c2ab62cf1dfa4b78559?d=retro&size=100"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2607/2607.29241.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2607.29241 in a model README.md to link it from this page.

Cite arxiv.org/abs/2607.29241 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2607.29241 in a Space README.md to link it from this page.

Add this paper to a collection to link it from this page.