{"id":"0c67e1a0b03a9e9b2cd79f05be6f52fe069f10054836182a150142108f0529fc","title":"arXiv cs.AI paper: Taobao’s URED recycles overlooked candidates for discovery","argument":"Ge Fan and coauthors report that URED, a training-free recall-stage system, has served Taobao feed recommendation at Alibaba for over six months. They argue that boosting under-explored content can backfire: cold, information-poor candidates may be over-estimated by rankers, risking engagement without improving discovery. URED instead gathers items scored highly but never exposed, aggregates them over a 30-day horizon, and re-injects them at recall time. The authors report that this adds under 1% latency without model retraining. In online A/B tests, they report increases in discovered categories and content richness with engagement preserved. Completion video views rose significantly; total views moved positive, but the increase was not significant. Their ablations report that a one-day window sharply reduced the discovery gain, while random injection without score filtering yielded no gain and trended negatively on engagement. The authors share deployment details and lessons for practitioners facing the exploration-exploitation dilemma under tight resource constraints.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence is the arXiv abstract, not the paper’s full methods or results; it does not provide the online A/B test design or sample sizes. The reported test outcomes are the authors’ claims; the supplied abstract does not provide enough detail to assess their statistical methods or independently evaluate the results.","speaker":{"name":"Fan, Ge, Meng, Kai, Chu, Huiping, Jiang, Yuning, Zhu, Han, Zheng, Bo","roleAtSource":"Article author"},"source":{"title":"From Waste to Discovery: Recycling High-Scored Unexposed Candidates in Taobao Feed Recommendation","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.05997","publishedAt":"2026-11-04","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors report that URED has served Taobao feed recommendation at Alibaba for over six months and is training-free.","excerpt":"a training-free recall-stage system serving Taobao feed recommendation at Alibaba for over six months","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"},{"text":"The authors argue that naive exposure boosting can inject cold, information-poor candidates that rankers over-estimate, risking engagement without improving discovery.","excerpt":"naive boosting fails because it injects cold, information-poor candidates that rankers systematically over-estimate, buying no discovery gain while risking engagement","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"},{"text":"URED aggregates highly scored items that have never been exposed over a 30-day horizon, then re-injects them at recall time.","excerpt":"Items scored highly but never exposed (signals industrial pipelines routinely discard) form a pool that is both relevant and reliably predictable. Over a 30-day horizon URED aggregates them","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"},{"text":"The authors report that URED adds under 1% latency and requires no model retraining.","excerpt":"adding under 1% latency via a single O(1) lookup per request with no model retraining.","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"},{"text":"In online A/B tests, the authors report increases in discovered categories and content richness, preserved engagement, and a significant rise in completion video views; total views moved positive but not significantly.","excerpt":"Online A/B tests show +3.47% discovered categories and +8.47% content richness with engagement preserved: completion video views rise significantly (+2.62%) and total views move positive (+0.36%, not significant).","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"},{"text":"The authors report that a one-day window sharply reduced the discovery gain, while random injection without score filtering yielded no gain and trended negatively on engagement.","excerpt":"a one-day window collapses the discovery gain to +0.55% (not significant), and random injection without score filtering yields no gain while trending negative on engagement (-0.17%).","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.05997","checkedAt":"2026-11-10T04:08:09.215Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How were “discovered categories” and “content richness” defined and measured in the online A/B tests?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI paper: Taobao’s URED recycles overlooked candidates for discovery","summary":"Ge Fan and coauthors report that URED, a training-free recall-stage system, has served Taobao feed recommendation at Alibaba for over six months. They argue that boosting under-explored content can backfire: cold, information-poor candidates may be over-estimated by rankers, risking engagement without improving discovery. URED instead gathers items scored highly but never exposed, aggregates them over a 30-day horizon, and re-injects them at recall time. The authors report that this adds under 1% latency without model retraining. In online A/B tests, they report increases in discovered categories and content richness with engagement preserved. Completion video views rose significantly; total views moved positive, but the increase was not significant. Their ablations report that a one-day window sharply reduced the discovery gain, while random injection without score filtering yielded no gain and trended negatively on engagement. The authors share deployment details and lessons for practitioners facing the exploration-exploitation dilemma under tight resource constraints.","question":"How were “discovered categories” and “content richness” defined and measured in the online A/B tests?","claims":[{"text":"The authors report that URED has served Taobao feed recommendation at Alibaba for over six months and is training-free.","excerpt":"a training-free recall-stage system serving Taobao feed recommendation at Alibaba for over six months","support":"source-stated"},{"text":"The authors argue that naive exposure boosting can inject cold, information-poor candidates that rankers over-estimate, risking engagement without improving discovery.","excerpt":"naive boosting fails because it injects cold, information-poor candidates that rankers systematically over-estimate, buying no discovery gain while risking engagement","support":"source-stated"},{"text":"URED aggregates highly scored items that have never been exposed over a 30-day horizon, then re-injects them at recall time.","excerpt":"Items scored highly but never exposed (signals industrial pipelines routinely discard) form a pool that is both relevant and reliably predictable. Over a 30-day horizon URED aggregates them","support":"source-stated"},{"text":"The authors report that URED adds under 1% latency and requires no model retraining.","excerpt":"adding under 1% latency via a single O(1) lookup per request with no model retraining.","support":"source-stated"},{"text":"In online A/B tests, the authors report increases in discovered categories and content richness, preserved engagement, and a significant rise in completion video views; total views moved positive but not significantly.","excerpt":"Online A/B tests show +3.47% discovered categories and +8.47% content richness with engagement preserved: completion video views rise significantly (+2.62%) and total views move positive (+0.36%, not significant).","support":"source-stated"},{"text":"The authors report that a one-day window sharply reduced the discovery gain, while random injection without score filtering yielded no gain and trended negatively on engagement.","excerpt":"a one-day window collapses the discovery gain to +0.55% (not significant), and random injection without score filtering yields no gain while trending negative on engagement (-0.17%).","support":"source-stated"}],"limitations":["The supplied evidence is the arXiv abstract, not the paper’s full methods or results; it does not provide the online A/B test design or sample sizes.","The reported test outcomes are the authors’ claims; the supplied abstract does not provide enough detail to assess their statistical methods or independently evaluate the results."],"mediaId":"632ce549c912c6915a202a5e7473104200c042cb12a75bca0ac0818d8791a694","correctionReason":null,"id":"0c67e1a0b03a9e9b2cd79f05be6f52fe069f10054836182a150142108f0529fc","publicationId":"7404fde5-0805-4833-8116-53fec5137284","version":1,"sourceUrl":"https://arxiv.org/abs/2611.05997","sourceTitle":"From Waste to Discovery: Recycling High-Scored Unexposed Candidates in Taobao Feed Recommendation","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-04","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-05T05:06:22.134Z","createdAt":"2026-11-09T11:04:24.983Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"632ce549c912c6915a202a5e7473104200c042cb12a75bca0ac0818d8791a694","kind":"article","url":"https://arxiv.org/abs/2611.05997","sourceUrl":"https://arxiv.org/abs/2611.05997","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T11:03:32.662Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.05997","checkedAt":"2026-11-10T04:08:09.215Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"632ce549c912c6915a202a5e7473104200c042cb12a75bca0ac0818d8791a694","confidence":0.8200000000000001,"provider":"typesafe","model":"jev-1.13.0","inputHash":"67291592b3248fa0477b7e946241c2c2d1d8a77493dffc895d6c7288da3b715e","decidedAt":"2026-11-09T11:03:34.594Z"},"provenance":{"textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observationId":"65b5e3b9558c97d50bdd339d45ee3032e9724091d5bb97d9269291598027aeb9","retrievedAt":"2026-11-09T11:03:32.353Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"feb5f1fe-113d-47d4-8e2f-0733c4f4959b"},"reviewer":{"model":"gpt-6-luna","runId":"e29fb4e0-943c-48c1-86f8-d1b36583d13b"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-09T11:04:24.983Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Fan, Ge","url":null,"kind":"unknown"},{"name":"Meng, Kai","url":null,"kind":"unknown"},{"name":"Chu, Huiping","url":null,"kind":"unknown"},{"name":"Jiang, Yuning","url":null,"kind":"unknown"},{"name":"Zhu, Han","url":null,"kind":"unknown"},{"name":"Zheng, Bo","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Fan, Ge","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Fan, Ge","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}},{"actor":{"name":"Meng, Kai","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Meng, Kai","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}},{"actor":{"name":"Chu, Huiping","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Chu, Huiping","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}},{"actor":{"name":"Jiang, Yuning","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Jiang, Yuning","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}},{"actor":{"name":"Zhu, Han","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhu, Han","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}},{"actor":{"name":"Zheng, Bo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zheng, Bo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}}],"participationCoverage":{"total":6,"returned":6,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.05997","snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/0c67e1a0b03a9e9b2cd79f05be6f52fe069f10054836182a150142108f0529fc/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-paper-taobao-s-ured-recycles-overlooked-candidates-0c67e1a0","formats":{"html":"/readings/arxiv-cs-ai-paper-taobao-s-ured-recycles-overlooked-candidates-0c67e1a0","markdown":"/readings/arxiv-cs-ai-paper-taobao-s-ured-recycles-overlooked-candidates-0c67e1a0.md","json":"/readings/arxiv-cs-ai-paper-taobao-s-ured-recycles-overlooked-candidates-0c67e1a0.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-latency","name":"Latency","explanation":"The delay between sending a request and receiving the result.","referenceUrl":"https://developers.google.com/machine-learning/glossary#latency"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"},{"id":"time","name":"TIME","kind":"organization","context":"TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.","referenceUrls":["https://time.com/"],"path":"/entities/time","json":"/entities/time.json","markdown":"/entities/time.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"0c67e1a0b03a9e9b2cd79f05be6f52fe069f10054836182a150142108f0529fc","revision":"7404fde5-0805-4833-8116-53fec5137284","revisionKind":"publication-id","title":"arXiv cs.AI paper: Taobao’s URED recycles overlooked candidates for discovery","url":"https://escapemeridian.com/readings/arxiv-cs-ai-paper-taobao-s-ured-recycles-overlooked-candidates-0c67e1a0"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.05997","revision":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","revisionKind":"sha256-source-text","title":"From Waste to Discovery: Recycling High-Scored Unexposed Candidates in Taobao Feed Recommendation","url":"https://arxiv.org/abs/2611.05997"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"0c67e1a0b03a9e9b2cd79f05be6f52fe069f10054836182a150142108f0529fc","title":"arXiv cs.AI paper: Taobao’s URED recycles overlooked candidates for discovery","argument":"Ge Fan and coauthors report that URED, a training-free recall-stage system, has served Taobao feed recommendation at Alibaba for over six months. They argue that boosting under-explored content can backfire: cold, information-poor candidates may be over-estimated by rankers, risking engagement without improving discovery. URED instead gathers items scored highly but never exposed, aggregates them over a 30-day horizon, and re-injects them at recall time. The authors report that this adds under 1% latency without model retraining. In online A/B tests, they report increases in discovered categories and content richness with engagement preserved. Completion video views rose significantly; total views moved positive, but the increase was not significant. Their ablations report that a one-day window sharply reduced the discovery gain, while random injection without score filtering yielded no gain and trended negatively on engagement. The authors share deployment details and lessons for practitioners facing the exploration-exploitation dilemma under tight resource constraints.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence is the arXiv abstract, not the paper’s full methods or results; it does not provide the online A/B test design or sample sizes. The reported test outcomes are the authors’ claims; the supplied abstract does not provide enough detail to assess their statistical methods or independently evaluate the results.","speaker":{"name":"Fan, Ge, Meng, Kai, Chu, Huiping, Jiang, Yuning, Zhu, Han, Zheng, Bo","roleAtSource":"Article author"},"source":{"title":"From Waste to Discovery: Recycling High-Scored Unexposed Candidates in Taobao Feed Recommendation","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.05997","publishedAt":"2026-11-04","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors report that URED has served Taobao feed recommendation at Alibaba for over six months and is training-free.","excerpt":"a training-free recall-stage system serving Taobao feed recommendation at Alibaba for over six months","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"},{"text":"The authors argue that naive exposure boosting can inject cold, information-poor candidates that rankers over-estimate, risking engagement without improving discovery.","excerpt":"naive boosting fails because it injects cold, information-poor candidates that rankers systematically over-estimate, buying no discovery gain while risking engagement","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"},{"text":"URED aggregates highly scored items that have never been exposed over a 30-day horizon, then re-injects them at recall time.","excerpt":"Items scored highly but never exposed (signals industrial pipelines routinely discard) form a pool that is both relevant and reliably predictable. Over a 30-day horizon URED aggregates them","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"},{"text":"The authors report that URED adds under 1% latency and requires no model retraining.","excerpt":"adding under 1% latency via a single O(1) lookup per request with no model retraining.","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"},{"text":"In online A/B tests, the authors report increases in discovered categories and content richness, preserved engagement, and a significant rise in completion video views; total views moved positive but not significantly.","excerpt":"Online A/B tests show +3.47% discovered categories and +8.47% content richness with engagement preserved: completion video views rise significantly (+2.62%) and total views move positive (+0.36%, not significant).","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"},{"text":"The authors report that a one-day window sharply reduced the discovery gain, while random injection without score filtering yielded no gain and trended negatively on engagement.","excerpt":"a one-day window collapses the discovery gain to +0.55% (not significant), and random injection without score filtering yields no gain while trending negative on engagement (-0.17%).","url":"https://arxiv.org/abs/2611.05997","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.05997","checkedAt":"2026-11-10T04:08:09.215Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How were “discovered categories” and “content richness” defined and measured in the online A/B tests?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI paper: Taobao’s URED recycles overlooked candidates for discovery","summary":"Ge Fan and coauthors report that URED, a training-free recall-stage system, has served Taobao feed recommendation at Alibaba for over six months. They argue that boosting under-explored content can backfire: cold, information-poor candidates may be over-estimated by rankers, risking engagement without improving discovery. URED instead gathers items scored highly but never exposed, aggregates them over a 30-day horizon, and re-injects them at recall time. The authors report that this adds under 1% latency without model retraining. In online A/B tests, they report increases in discovered categories and content richness with engagement preserved. Completion video views rose significantly; total views moved positive, but the increase was not significant. Their ablations report that a one-day window sharply reduced the discovery gain, while random injection without score filtering yielded no gain and trended negatively on engagement. The authors share deployment details and lessons for practitioners facing the exploration-exploitation dilemma under tight resource constraints.","question":"How were “discovered categories” and “content richness” defined and measured in the online A/B tests?","claims":[{"text":"The authors report that URED has served Taobao feed recommendation at Alibaba for over six months and is training-free.","excerpt":"a training-free recall-stage system serving Taobao feed recommendation at Alibaba for over six months","support":"source-stated"},{"text":"The authors argue that naive exposure boosting can inject cold, information-poor candidates that rankers over-estimate, risking engagement without improving discovery.","excerpt":"naive boosting fails because it injects cold, information-poor candidates that rankers systematically over-estimate, buying no discovery gain while risking engagement","support":"source-stated"},{"text":"URED aggregates highly scored items that have never been exposed over a 30-day horizon, then re-injects them at recall time.","excerpt":"Items scored highly but never exposed (signals industrial pipelines routinely discard) form a pool that is both relevant and reliably predictable. Over a 30-day horizon URED aggregates them","support":"source-stated"},{"text":"The authors report that URED adds under 1% latency and requires no model retraining.","excerpt":"adding under 1% latency via a single O(1) lookup per request with no model retraining.","support":"source-stated"},{"text":"In online A/B tests, the authors report increases in discovered categories and content richness, preserved engagement, and a significant rise in completion video views; total views moved positive but not significantly.","excerpt":"Online A/B tests show +3.47% discovered categories and +8.47% content richness with engagement preserved: completion video views rise significantly (+2.62%) and total views move positive (+0.36%, not significant).","support":"source-stated"},{"text":"The authors report that a one-day window sharply reduced the discovery gain, while random injection without score filtering yielded no gain and trended negatively on engagement.","excerpt":"a one-day window collapses the discovery gain to +0.55% (not significant), and random injection without score filtering yields no gain while trending negative on engagement (-0.17%).","support":"source-stated"}],"limitations":["The supplied evidence is the arXiv abstract, not the paper’s full methods or results; it does not provide the online A/B test design or sample sizes.","The reported test outcomes are the authors’ claims; the supplied abstract does not provide enough detail to assess their statistical methods or independently evaluate the results."],"mediaId":"632ce549c912c6915a202a5e7473104200c042cb12a75bca0ac0818d8791a694","correctionReason":null,"id":"0c67e1a0b03a9e9b2cd79f05be6f52fe069f10054836182a150142108f0529fc","publicationId":"7404fde5-0805-4833-8116-53fec5137284","version":1,"sourceUrl":"https://arxiv.org/abs/2611.05997","sourceTitle":"From Waste to Discovery: Recycling High-Scored Unexposed Candidates in Taobao Feed Recommendation","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-04","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-05T05:06:22.134Z","createdAt":"2026-11-09T11:04:24.983Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"632ce549c912c6915a202a5e7473104200c042cb12a75bca0ac0818d8791a694","kind":"article","url":"https://arxiv.org/abs/2611.05997","sourceUrl":"https://arxiv.org/abs/2611.05997","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T11:03:32.662Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.05997","checkedAt":"2026-11-10T04:08:09.215Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"632ce549c912c6915a202a5e7473104200c042cb12a75bca0ac0818d8791a694","confidence":0.8200000000000001,"provider":"typesafe","model":"jev-1.13.0","inputHash":"67291592b3248fa0477b7e946241c2c2d1d8a77493dffc895d6c7288da3b715e","decidedAt":"2026-11-09T11:03:34.594Z"},"provenance":{"textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observationId":"65b5e3b9558c97d50bdd339d45ee3032e9724091d5bb97d9269291598027aeb9","retrievedAt":"2026-11-09T11:03:32.353Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"feb5f1fe-113d-47d4-8e2f-0733c4f4959b"},"reviewer":{"model":"gpt-6-luna","runId":"e29fb4e0-943c-48c1-86f8-d1b36583d13b"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-09T11:04:24.983Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Fan, Ge","url":null,"kind":"unknown"},{"name":"Meng, Kai","url":null,"kind":"unknown"},{"name":"Chu, Huiping","url":null,"kind":"unknown"},{"name":"Jiang, Yuning","url":null,"kind":"unknown"},{"name":"Zhu, Han","url":null,"kind":"unknown"},{"name":"Zheng, Bo","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Fan, Ge","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Fan, Ge","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}},{"actor":{"name":"Meng, Kai","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Meng, Kai","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}},{"actor":{"name":"Chu, Huiping","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Chu, Huiping","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}},{"actor":{"name":"Jiang, Yuning","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Jiang, Yuning","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}},{"actor":{"name":"Zhu, Han","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhu, Han","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}},{"actor":{"name":"Zheng, Bo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zheng, Bo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05997","locator":"head/meta[name=author|citation_author]"},"assertionId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T04:08:09.215Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","sourceUrl":"https://arxiv.org/abs/2611.05997","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z"}}],"participationCoverage":{"total":6,"returned":6,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.05997","snapshotId":"d7a70c72b5f0cf83a88f256a8627ab9088817294b3d9421ffd3ce7b3053d343c","contentSha256":"8a8d0b9f35a13643d911e7ab6fb2b501913ceb2ae3bf2db0b5d3dabf333d66f0","textHash":"83951c1fb90edce64e6d5ffee13251c6601e3bfc3dc4243b6ffbbd5722ad1338","observedAt":"2026-11-10T04:08:09.215Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/0c67e1a0b03a9e9b2cd79f05be6f52fe069f10054836182a150142108f0529fc/source"}}}],"audits":[{"publicationId":"7404fde5-0805-4833-8116-53fec5137284","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes URED and its results to Ge Fan and coauthors, matching the arXiv author list. Results are consistently framed as what the authors report or argue, not as established fact.","excerpt":"Authors:Ge Fan, Kai Meng, Huiping Chu, Yuning Jiang, Han Zhu, Bo Zheng"},{"dimension":"summary","status":"uncertain","explanation":"The reading is faithful, but central jargon is unexplained for a non-specialist: recall-stage/recall time, cold candidates, rankers, ablations, and the exploration-exploitation dilemma. A general reader may not grasp where URED sits in the pipeline or why unexposed high scorers matter. It also omits the scale (tens of millions of daily users) and the reported 13.9x platform-average discovery rate, which convey significance. Smallest repairs: add plain-language glosses or term-library tooltips (for example, recall = the first step that shortlists candidate items before ranking), and add one sentence on scale. The title and question are reasonable.","excerpt":"serving Taobao feed recommendation at Alibaba for over six months to tens of millions of daily active users"},{"dimension":"date","status":"pass","explanation":"originalPublished is 2026-11-04, which matches the v1 submission date on arXiv. The feed date of 2026-11-05 is correctly not used.","excerpt":"[Submitted on 4 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim is directly supported by the abstract. 'Sharply reduced' fairly renders 'collapses' to +0.55% (not significant). Significance caveats on total views are preserved. The limitations correctly note that only the abstract is available. No unsupported benefits or costs are implied.","excerpt":"a one-day window collapses the discovery gain to +0.55% (not significant)"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the most direct access to the source.","excerpt":null}],"createdAt":"2026-11-09T11:04:24.983Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}