[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-llms-recommend-badly-when-item-descriptions-vary-in-length":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},3930,"llms-recommend-badly-when-item-descriptions-vary-in-length","LLMs Recommend Badly When Item Descriptions Vary in Length","A new framework called LBR targets a quiet flaw in AI recommendation systems: longer product descriptions skew results in two opposite directions at once.","AI recommendation systems built on large language models have a length problem nobody was talking about.\n\nResearchers have identified what they call \"length bias\" in LLM-based recommenders — systems that treat recommendation as a text generation task. The bias runs in two directions simultaneously. On the input side, items with longer descriptions grab more attention from the model simply because they consume more tokens, distorting how the model weighs user preferences. On the output side, the same long items get penalized when the model scores candidates by summed log-likelihood, a method that structurally disadvantages longer text. The obvious fix — normalizing by token count — turns out to make things worse, not better.\n\nThe proposed remedy, a framework called LBR (Length Bias Reduction), attacks both problems without retraining the underlying model. It adjusts attention scores during input processing using a length-dependent offset, and replaces naive token-count normalization with a measure derived from the branching structure of a prefix tree — a more information-theoretic way to gauge how much a description actually says versus how many words it uses. Tested across three Amazon product datasets, LBR produced an average 16.82% gain on a standard ranking metric (NDCG@5) with minimal added compute.\n\nThe finding matters beyond academic benchmarks. Most commercial recommendation pipelines are increasingly stitched together with LLMs, and product catalogs are notoriously uneven — a flagship phone gets a 500-word description; a USB cable gets ten words. That asymmetry alone could be quietly distorting what surfaces at the top of results. LBR is model-agnostic and the code is public, which lowers the bar for adoption — though whether product teams bother to retrofit it is another question entirely.","[\"ai\",\"machine-learning\",\"recommendation-systems\",\"llm\"]","2026-07-07T04:00:00.000Z","2026-07-07T12:23:54.882Z","2026-07-07T12:23:57.798Z","published",null,[],"ai",[24,26,27,28],"machine-learning","recommendation-systems","llm",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.04270",0,{"sections":35},[36,40,45,50,55,60,65,70,75,79,84,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":18},"Dev Tools","dev-tools",59,{"name":80,"slug":81,"count":82,"latest_published_at":83},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":85,"slug":86,"count":82,"latest_published_at":87},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]