让搜索“一步到位”！快手(01024)提出端到端生成式搜索方案OneSearch_老虎社区_美港股上老虎

让搜索“一步到位”！快手(01024)提出端到端生成式搜索方案OneSearch

智通财经2025-09-23

智通财经APP获悉，当前，电商平台普遍采用“召回、粗排、精排”的级联式搜索架构。该架构虽然成熟稳定，但仍面临诸多痛点：商品描述混乱、相关性问题突出、级联结构存在瓶颈以及冷启动难题，导致搜索结果往往不尽如人意。为了解决这些挑战，快手(01024)提出了业界首个工业级部署的电商搜索端到端生成式框架——OneSearch，目前，该系统已在快手多个电商搜索场景中成功部署，每日服务数百万用户，产生数千万页面...

网页链接

免责声明：本文观点仅代表作者个人观点，不构成本平台的投资建议，本平台不对文章信息准确性、完整性和及时性做出任何保证，亦不对因使用或信赖文章信息引发的任何损失承担责任。

湿地蝙蝠侠
·2025-09-23
快手的ai应用很快
回复
2
举报

热议股票

7x24

全部

{"i18n":{"language":"zh_CN"},"data":{"share":"https://www.laohu8.com/m/news/2569979252?lang=zh_CN&edition=full","thumbnail":"","is_english":false,"pubTime":"2025-09-23 19:18","share_image_url":"https://static.laohu8.com/e9f99090a1c2ed51c021029395664489","id":"2569979252","market":"fut","top_or_hot":-1,"title":"让搜索“一步到位”！快手(01024)提出端到端生成式搜索方案OneSearch","media":"智通财经","content":"<div>\n<p>智通财经APP获悉，当前，电商平台普遍采用“召回、粗排、精排”的级联式搜索架构。该架构虽然成熟稳定，但仍面临诸多痛点：商品描述混乱、相关性问题突出、级联结构存在瓶颈以及冷启动难题，导致搜索结果往往不尽如人意。为了解决这些挑战，快手(01024)提出了业界首个工业级部署的电商搜索端到端生成式框架——OneSearch，目前，该系统已在快手多个电商搜索场景中成功部署，每日服务数百万用户，产生数千万页面...</p>\n\n<a href=\"http://www.zhitongcaijing.com/content/detail/1348592.html\">网页链接</a>\n\n</div>\n","source":"stock_zhitongcaijing","html":"<!DOCTYPE html>\n<html>\n<head>\n<meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />\n<meta name=\"viewport\" content=\"width=device-width,initial-scale=1.0,minimum-scale=1.0,maximum-scale=1.0,user-scalable=no\"/>\n<meta name=\"format-detection\" content=\"telephone=no,email=no,address=no\" />\n<title>让搜索“一步到位”！快手(01024)提出端到端生成式搜索方案OneSearch</title>\n<style type=\"text/css\">\na,abbr,acronym,address,applet,article,aside,audio,b,big,blockquote,body,canvas,caption,center,cite,code,dd,del,details,dfn,div,dl,dt,\nem,embed,fieldset,figcaption,figure,footer,form,h1,h2,h3,h4,h5,h6,header,hgroup,html,i,iframe,img,ins,kbd,label,legend,li,mark,menu,nav,\nobject,ol,output,p,pre,q,ruby,s,samp,section,small,span,strike,strong,sub,summary,sup,table,tbody,td,tfoot,th,thead,time,tr,tt,u,ul,var,video{ font:inherit;margin:0;padding:0;vertical-align:baseline;border:0 }\nbody{ font-size:16px; line-height:1.5; color:#999; background:transparent; }\n.wrapper{ overflow:hidden;word-break:break-all;padding:10px; }\nh1,h2{ font-weight:normal; line-height:1.35; margin-bottom:.6em; }\nh3,h4,h5,h6{ line-height:1.35; margin-bottom:1em; }\nh1{ font-size:24px; }\nh2{ font-size:20px; }\nh3{ font-size:18px; }\nh4{ font-size:16px; }\nh5{ font-size:14px; }\nh6{ font-size:12px; }\np,ul,ol,blockquote,dl,table{ margin:1.2em 0; }\nul,ol{ margin-left:2em; }\nul{ list-style:disc; }\nol{ list-style:decimal; }\nli,li p{ margin:10px 0;}\nimg{ max-width:100%;display:block;margin:0 auto 1em; }\nblockquote{ color:#B5B2B1; border-left:3px solid #aaa; padding:1em; }\nstrong,b{font-weight:bold;}\nem,i{font-style:italic;}\ntable{ width:100%;border-collapse:collapse;border-spacing:1px;margin:1em 0;font-size:.9em; }\nth,td{ padding:5px;text-align:left;border:1px solid #aaa; }\nth{ font-weight:bold;background:#5d5d5d; }\n.symbol-link{font-weight:bold;}\n/* header{ border-bottom:1px solid #494756; } */\n.title{ margin:0 0 8px;line-height:1.3;color:#ddd; }\n.meta {color:#5e5c6d;font-size:13px;margin:0 0 .5em; }\na{text-decoration:none; color:#2a4b87;}\n.meta .head { display: inline-block; overflow: hidden}\n.head .h-thumb { width: 30px; height: 30px; margin: 0; padding: 0; border-radius: 50%; float: left;}\n.head .h-content { margin: 0; padding: 0 0 0 9px; float: left;}\n.head .h-name {font-size: 13px; color: #eee; margin: 0;}\n.head .h-time {font-size: 11px; color: #7E829C; margin: 0;line-height: 11px;}\n.small {font-size: 12.5px; display: inline-block; transform: scale(0.9); -webkit-transform: scale(0.9); transform-origin: left; -webkit-transform-origin: left;}\n.smaller {font-size: 12.5px; display: inline-block; transform: scale(0.8); -webkit-transform: scale(0.8); transform-origin: left; -webkit-transform-origin: left;}\n.bt-text {font-size: 12px;margin: 1.5em 0 0 0}\n.bt-text p {margin: 0}\n</style>\n</head>\n<body>\n<div class=\"wrapper\">\n<header>\n<h2 class=\"title\">\n让搜索“一步到位”！快手(01024)提出端到端生成式搜索方案OneSearch\n</h2>\n\n<h4 class=\"meta\">\n\n\n2025-09-23 19:18 北京时间&nbsp;&nbsp;&nbsp;<a href=http://www.zhitongcaijing.com/content/detail/1348592.html><strong>智通财经</strong></a>\n\n\n</h4>\n\n</header>\n<article>\n<div>\n<p>智通财经APP获悉，当前，电商平台普遍采用“召回、粗排、精排”的级联式搜索架构。该架构虽然成熟稳定，但仍面临诸多痛点：商品描述混乱、相关性问题突出、级联结构存在瓶颈以及冷启动难题，导致搜索结果往往不尽如人意。为了解决这些挑战，快手(01024)提出了业界首个工业级部署的电商搜索端到端生成式框架——OneSearch，目前，该系统已在快手多个电商搜索场景中成功部署，每日服务数百万用户，产生数千万页面...</p>\n\n<a href=\"http://www.zhitongcaijing.com/content/detail/1348592.html\">网页链接</a>\n\n</div>\n\n\n</article>\n</div>\n</body>\n</html>\n","isBrief":false,"type":0,"news_type":1,"symbol":"LU2097828631.EUR","symbol_name":"AZ EQUITY CHINA \"A\" (EUR) ACC","start_time":0,"source_url":"http://www.zhitongcaijing.com/content/detail/1348592.html","article_id":"2569979252","we_media_id":null,"thumbnails":[],"rights":null,"url":"https://stock-news.laohu8.com/highlight/detail?id=2569979252","pubTimestamp":1758626337,"columns":[],"sourceInfo":{"source_id":"stock_zhitongcaijing","name":"智通财经网"},"weMediaInfo":null,"summary":"为了解决这些挑战，快手提出了业界首个工业级部署的电商搜索端到端生成式框架——OneSearch，目前，该系统已在快手多个电商搜索场景中成功部署，每日服务数百万用户，产生数千万页面浏览量。订单量、买家数等多项指标实现大幅提升离线实验表明，OneSearch相比现有级联式系统各项指标均有显著提升。在人工评估中，OneSearch不仅在CVR和CTR上表现优异，还在页面整体满意度、商品质量及query-item相关性方面均显著优于传统级联系统。","collect":0,"end_time":0,"defaultTopTitle":"zhitongcaijing.com","property":[],"viewcount":null,"language":"zh","relate_stocks":{"81024":"快手-WR","LU2097828631.EUR":"AZ EQUITY CHINA \"A\" (EUR) ACC","BK1575":"同股不同权","LU1794554557.SGD":"Allianz All China Equity AT Acc H2-SGD","LU1770034418.SGD":"ALL CHINA EQUITY \"A\" (SGDHDG) ACC","LU2097828714.EUR":"AZ EQUITY CHINA \"BAZ\" (EUR) ACC","LU0463099449.HKD":"SCHRODER ISF CHINA OPPORTUNITIES \"A\" (HKD) ACC","LU2097828805.USD":"AZ EQUITY CHINA \"A-AZ\" (USD) ACC","BK1591":"就地过年概念","BK1095":"互动媒体与服务","LU0348766576.USD":"ALLIANZ LITTLE DRAGONS \"A\" (USD) INC","LU0359201885.HKD":"BGF CHINA \"A2\" (HKD) ACC","LU0211977185.USD":"EASTSPRING INVESTMENTS GREATER CHINA EQUITY \"A\" ACC","BK1590":"短视频概念股","LU0140636845.USD":"施罗德大中华区股票A Acc","LU0348767384.USD":"ALLIANZLITTLE DRAGONS \"A\" (USD) ACC","LU0307460666.USD":"EASTSPRING INVESTMENTS CHINA EQUITY \"A\" ACC","LU2097828474.EUR":"AZ EQUITY CHINA \"A\" (EUR) ACC A","BK1615":"港股-互联网","LU0588546209.SGD":"Eastspring Investments - China Equity Fund AS SGD","LU0593848301.USD":"未来资产亚洲卓越消费股票基金A","LU1188198961.HKD":"SCHRODER ISF CHINA OPPORTUNITIES \"A\" (HKD) INC QV","LU0359201612.USD":"贝莱德中国基金A2","LU0244354667.USD":"SCHRODER ISF CHINA OPPORTUNITIES \"A\" ACC","LU1251922891.USD":"NINETY ONE GSF ALL CHINA EQUITY \"A\" (USD) ACC","LU0359202008.SGD":"Blackrock China Fund A2 SGD-H","LU0051755006.USD":"摩根大通中国A (dist)","01024":"快手-W","LU1720050803.USD":"安联全方位中国股票基金","LU1023057109.AUD":"BGF CHINA \"A2\" (AUDHDG) ACC","LU0210526637.USD":"JPM CHINA \"A\" (USD) ACC","LU0326950275.SGD":"Schroder ISF China Opportunities A Acc SGD-H","LU0456827905.SGD":"JPMorgan Funds - China A (acc) SGD","LU1303224171.USD":"NINETY ONE GSF ALL CHINA EQUITY \"A\" (USD) INC","BK1610":"ETF&股票定投概念","LU2097828557.USD":"AZ EQUITY CHINA \"A\" (USD) ACC","LU1719994722.HKD":"NINETY ONE GSF ALL CHINA EQUITY \"A\" (HKD) ACC","BK1608":"元宇宙概念"},"translate_title":"Make the search \"in one step\"! Kuaishou (01024) proposes the end-to-end generative search solution OneSearch","themeId":null,"isJumpTheme":false,"ttsUrl":null,"symbols_score_info":{"81024":0.6,"01024":1.5,"KSTmain":1.5},"content_text":"智通财经APP获悉，当前，电商平台普遍采用“召回、粗排、精排”的级联式搜索架构。该架构虽然成熟稳定，但仍面临诸多痛点：商品描述混乱、相关性问题突出、级联结构存在瓶颈以及冷启动难题，导致搜索结果往往不尽如人意。为了解决这些挑战，快手(01024)提出了业界首个工业级部署的电商搜索端到端生成式框架——OneSearch，目前，该系统已在快手多个电商搜索场景中成功部署，每日服务数百万用户，产生数千万页面浏览量。打破传统架构 提出创新解决方案OneSearch框架集三大创新于一身：关键词增强层次量化编码(KHQE)模块、多视角用户行为序列注入策略以及偏好感知奖励系统(PARS)。在关键词增强层次量化编码(KHQE)模块中，采用RQ-OPQ编码方案，从纵向与横向两个维度建模商品特征，为每个商品生成具备丰富语义层次的“智能身份证”，极大提升生成式检索的区分能力和准确性。多视角用户行为序列注入策略则让OneSearch能有效捕捉用户的近期偏好与长期兴趣，基于用户的长短期行为序列构建具有区分性的用户标识。使系统实现更全面、深层的用户意图理解，显著提升个性化搜索准确性与用户体验。偏好感知奖励系统(PARS)是结合多阶段监督微调与自适应奖励强化学习机制，捕捉细粒度用户偏好信号。该机制在提升排序性能的同时，保障生成多样性，有效避免“奖励破解”问题。订单量、买家数等多项指标实现大幅提升离线实验表明，OneSearch相比现有级联式系统各项指标均有显著提升。在线部署结果更为突出：订单量提升3.22%，买家数增长2.4%。这是在大规模工业场景下，这是生成式模型首次在大规模工业场景中完整替代传统搜索链路，具备重要落地意义。在人工评估中，OneSearch不仅在CVR和CTR上表现优异，还在页面整体满意度、商品质量及query-item相关性方面均显著优于传统级联系统。此外，系统在线性能同样亮眼：机器计算效率(MFU)相对提升达8倍，线上推理成本(OPEX)降低75.40%，资源利用效率得到极大优化。特别值得关注的是，OneSearch在冷启动场景下表现尤为突出，效果明显优于常规场景，表面生成式检索模型能够有效应对长尾用户和新上架商品的排序挑战。OneSearch的成功部署，标志着生成式模型在大规模工业搜索场景中取代传统链路的重大突破，为电商搜索技术的未来发展指明了方向。相关团队多项技术突破已发表在RecSys、CIKM、KDD等国际顶级会议上，引起业界广泛关注。后续，快手将持续探索在线实时编码方案，缩小预定义编码与流式训练之间的差异。同时，还将引入更强大的强化学习机制以更精准地匹配用户偏好，并结合图像、视频等多模态商品特征，进一步提升模型的推理效果与用户体验。随着技术的不断迭代，未来的电商搜索将更加智能、精准和“懂你”，让用户真正实现“一步到位”的理想搜索体验。","kind":"news","is_publish_news":true,"is_publish_highlight":false,"is_publish_live":false,"is_publish_wemedia":null,"editions":null,"column":"","sentiment":"0","news_tag":"productRelease","news_rank":0,"isVideo":false,"video":null,"symbols":[],"gpt_button":0,"need_auth":false,"need_login_tip":false,"code":"91000000","status":"200"},"commentList":[{"id":481706704609720,"gmtCreate":1758628140541,"gmtModify":1758631631780,"author":{"id":"3491099108319346","authorId":"3491099108319346","name":"湿地蝙蝠侠","avatar":"https://static.tigerbbs.com/0be88d31d32d3d732218dc95b5fd470c","vip":1,"crmLevel":1,"crmLevelSwitch":0,"wearingBadges":[],"idStr":"3491099108319346","authorIdStr":"3491099108319346"},"htmlText":"快手的ai应用很快","listText":"快手的ai应用很快","text":"快手的ai应用很快","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":2,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/481706704609720","repostId":2569979252,"repostType":2,"repost":{"id":"2569979252","kind":"news","pubTimestamp":1758626337,"share":"https://www.laohu8.com/m/news/2569979252?lang=zh_CN&edition=full","pubTime":"2025-09-23 19:18","market":"fut","language":"zh","title":"让搜索“一步到位”！快手(01024)提出端到端生成式搜索方案OneSearch","url":"https://stock-news.laohu8.com/highlight/detail?id=2569979252","media":"智通财经","summary":"为了解决这些挑战，快手提出了业界首个工业级部署的电商搜索端到端生成式框架——OneSearch，目前，该系统已在快手多个电商搜索场景中成功部署，每日服务数百万用户，产生数千万页面浏览量。订单量、买家数等多项指标实现大幅提升离线实验表明，OneSearch相比现有级联式系统各项指标均有显著提升。在人工评估中，OneSearch不仅在CVR和CTR上表现优异，还在页面整体满意度、商品质量及query-item相关性方面均显著优于传统级联系统。","content":"<html><body><p>智通财经APP获悉，当前，电商平台普遍采用“召回、粗排、精排”的级联式搜索架构。该架构虽然成熟稳定，但仍面临诸多痛点：商品描述混乱、相关性问题突出、级联结构存在瓶颈以及冷启动难题，导致搜索结果往往不尽如人意。</p><p>为了解决这些挑战，<a href=\"https://laohu8.com/S/01024\">快手</a><span>(01024)</span>提出了业界首个工业级部署的电商搜索端到端生成式框架——OneSearch，目前，该系统已在快手多个电商搜索场景中成功部署，每日服务数百万用户，产生数千万页面浏览量。</p><p><strong>打破传统架构 提出创新解决方案</strong></p><p>OneSearch框架集三大创新于一身：关键词增强层次量化编码(KHQE)模块、多视角用户行为序列注入策略以及偏好感知奖励系统(PARS)。</p><p>在关键词增强层次量化编码(KHQE)模块中，采用RQ-OPQ编码方案，从纵向与横向两个维度建模商品特征，为每个商品生成具备丰富语义层次的“智能身份证”，极大提升生成式检索的区分能力和准确性。</p><p>多视角用户行为序列注入策略则让OneSearch能有效捕捉用户的近期偏好与长期兴趣，基于用户的长短期行为序列构建具有区分性的用户标识。使系统实现更全面、深层的用户意图理解，显著提升个性化搜索准确性与用户体验。</p><p>偏好感知奖励系统(PARS)是结合多阶段监督微调与自适应奖励强化学习机制，捕捉细粒度用户偏好信号。该机制在提升排序性能的同时，保障生成多样性，有效避免“奖励破解”问题。</p><p><img src=\"https://img.zhitongcaijing.com/image/20250923/1758626251757421.png?x-oss-process=image/format,jpg/quality,q_80\" title=\"1758626251757421.png\"/></p><p><strong>订单量、买家数等多项指标实现大幅提升</strong></p><p>离线实验表明，OneSearch相比现有级联式系统各项指标均有显著提升。在线部署结果更为突出：订单量提升3.22%，买家数增长2.4%。这是在大规模工业场景下，这是生成式模型首次在大规模工业场景中完整替代传统搜索链路，具备重要落地意义。</p><p><img src=\"https://img.zhitongcaijing.com/image/20250923/1758626279867470.png?x-oss-process=image/format,jpg/quality,q_80\" title=\"1758626279867470.png\"/></p><p>在人工评估中，OneSearch不仅在CVR和CTR上表现优异，还在页面整体满意度、商品质量及query-item相关性方面均显著优于传统级联系统。此外，系统在线性能同样亮眼：机器计算效率(MFU)相对提升达8倍，线上推理成本(OPEX)降低75.40%，资源利用效率得到极大优化。</p><p><img src=\"https://img.zhitongcaijing.com/image/20250923/1758626298363305.png?x-oss-process=image/format,jpg/quality,q_80\" title=\"1758626298363305.png\"/></p><p>特别值得关注的是，OneSearch在冷启动场景下表现尤为突出，效果明显优于常规场景，表面生成式检索模型能够有效应对长尾用户和新上架商品的排序挑战。</p><p>OneSearch的成功部署，标志着生成式模型在大规模工业搜索场景中取代传统链路的重大突破，为电商搜索技术的未来发展指明了方向。相关团队多项技术突破已发表在RecSys、CIKM、KDD等国际顶级会议上，引起业界广泛关注。</p><p>后续，快手将持续探索在线实时编码方案，缩小预定义编码与流式训练之间的差异。同时，还将引入更强大的强化学习机制以更精准地匹配用户偏好，并结合图像、视频等多模态商品特征，进一步提升模型的推理效果与用户体验。</p><p>随着技术的不断迭代，未来的电商搜索将更加智能、精准和“懂你”，让用户真正实现“一步到位”的<a href=\"https://laohu8.com/S/LI\">理想</a>搜索体验。</p></body></html>","source":"stock_zhitongcaijing","collect":0,"html":"<!DOCTYPE html>\n<html>\n<head>\n<meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />\n<meta name=\"viewport\" content=\"width=device-width,initial-scale=1.0,minimum-scale=1.0,maximum-scale=1.0,user-scalable=no\"/>\n<meta name=\"format-detection\" content=\"telephone=no,email=no,address=no\" />\n<title>让搜索“一步到位”！快手(01024)提出端到端生成式搜索方案OneSearch</title>\n<style type=\"text/css\">\na,abbr,acronym,address,applet,article,aside,audio,b,big,blockquote,body,canvas,caption,center,cite,code,dd,del,details,dfn,div,dl,dt,\nem,embed,fieldset,figcaption,figure,footer,form,h1,h2,h3,h4,h5,h6,header,hgroup,html,i,iframe,img,ins,kbd,label,legend,li,mark,menu,nav,\nobject,ol,output,p,pre,q,ruby,s,samp,section,small,span,strike,strong,sub,summary,sup,table,tbody,td,tfoot,th,thead,time,tr,tt,u,ul,var,video{ font:inherit;margin:0;padding:0;vertical-align:baseline;border:0 }\nbody{ font-size:16px; line-height:1.5; color:#999; background:transparent; }\n.wrapper{ overflow:hidden;word-break:break-all;padding:10px; }\nh1,h2{ font-weight:normal; line-height:1.35; margin-bottom:.6em; }\nh3,h4,h5,h6{ line-height:1.35; margin-bottom:1em; }\nh1{ font-size:24px; }\nh2{ font-size:20px; }\nh3{ font-size:18px; }\nh4{ font-size:16px; }\nh5{ font-size:14px; }\nh6{ font-size:12px; }\np,ul,ol,blockquote,dl,table{ margin:1.2em 0; }\nul,ol{ margin-left:2em; }\nul{ list-style:disc; }\nol{ list-style:decimal; }\nli,li p{ margin:10px 0;}\nimg{ max-width:100%;display:block;margin:0 auto 1em; }\nblockquote{ color:#B5B2B1; border-left:3px solid #aaa; padding:1em; }\nstrong,b{font-weight:bold;}\nem,i{font-style:italic;}\ntable{ width:100%;border-collapse:collapse;border-spacing:1px;margin:1em 0;font-size:.9em; }\nth,td{ padding:5px;text-align:left;border:1px solid #aaa; }\nth{ font-weight:bold;background:#5d5d5d; }\n.symbol-link{font-weight:bold;}\n/* header{ border-bottom:1px solid #494756; } */\n.title{ margin:0 0 8px;line-height:1.3;color:#ddd; }\n.meta {color:#5e5c6d;font-size:13px;margin:0 0 .5em; }\na{text-decoration:none; color:#2a4b87;}\n.meta .head { display: inline-block; overflow: hidden}\n.head .h-thumb { width: 30px; height: 30px; margin: 0; padding: 0; border-radius: 50%; float: left;}\n.head .h-content { margin: 0; padding: 0 0 0 9px; float: left;}\n.head .h-name {font-size: 13px; color: #eee; margin: 0;}\n.head .h-time {font-size: 11px; color: #7E829C; margin: 0;line-height: 11px;}\n.small {font-size: 12.5px; display: inline-block; transform: scale(0.9); -webkit-transform: scale(0.9); transform-origin: left; -webkit-transform-origin: left;}\n.smaller {font-size: 12.5px; display: inline-block; transform: scale(0.8); -webkit-transform: scale(0.8); transform-origin: left; -webkit-transform-origin: left;}\n.bt-text {font-size: 12px;margin: 1.5em 0 0 0}\n.bt-text p {margin: 0}\n</style>\n</head>\n<body>\n<div class=\"wrapper\">\n<header>\n<h2 class=\"title\">\n让搜索“一步到位”！快手(01024)提出端到端生成式搜索方案OneSearch\n</h2>\n\n<h4 class=\"meta\">\n\n\n2025-09-23 19:18 北京时间&nbsp;&nbsp;&nbsp;<a href=http://www.zhitongcaijing.com/content/detail/1348592.html><strong>智通财经</strong></a>\n\n\n</h4>\n\n</header>\n<article>\n<div>\n<p>智通财经APP获悉，当前，电商平台普遍采用“召回、粗排、精排”的级联式搜索架构。该架构虽然成熟稳定，但仍面临诸多痛点：商品描述混乱、相关性问题突出、级联结构存在瓶颈以及冷启动难题，导致搜索结果往往不尽如人意。为了解决这些挑战，快手(01024)提出了业界首个工业级部署的电商搜索端到端生成式框架——OneSearch，目前，该系统已在快手多个电商搜索场景中成功部署，每日服务数百万用户，产生数千万页面...</p>\n\n<a href=\"http://www.zhitongcaijing.com/content/detail/1348592.html\">网页链接</a>\n\n</div>\n\n\n</article>\n</div>\n</body>\n</html>\n","type":0,"thumbnail":"","relate_stocks":{"81024":"快手-WR","LU2097828631.EUR":"AZ EQUITY CHINA \"A\" (EUR) ACC","BK1575":"同股不同权","LU1794554557.SGD":"Allianz All China Equity AT Acc H2-SGD","LU1770034418.SGD":"ALL CHINA EQUITY \"A\" (SGDHDG) ACC","LU2097828714.EUR":"AZ EQUITY CHINA \"BAZ\" (EUR) ACC","LU0463099449.HKD":"SCHRODER ISF CHINA OPPORTUNITIES \"A\" (HKD) ACC","LU2097828805.USD":"AZ EQUITY CHINA \"A-AZ\" (USD) ACC","BK1591":"就地过年概念","BK1095":"互动媒体与服务","LU0348766576.USD":"ALLIANZ LITTLE DRAGONS \"A\" (USD) INC","LU0359201885.HKD":"BGF CHINA \"A2\" (HKD) ACC","LU0211977185.USD":"EASTSPRING INVESTMENTS GREATER CHINA EQUITY \"A\" ACC","BK1590":"短视频概念股","LU0140636845.USD":"施罗德大中华区股票A Acc","LU0348767384.USD":"ALLIANZLITTLE DRAGONS \"A\" (USD) ACC","LU0307460666.USD":"EASTSPRING INVESTMENTS CHINA EQUITY \"A\" ACC","LU2097828474.EUR":"AZ EQUITY CHINA \"A\" (EUR) ACC A","BK1615":"港股-互联网","LU0588546209.SGD":"Eastspring Investments - China Equity Fund AS SGD","LU0593848301.USD":"未来资产亚洲卓越消费股票基金A","LU1188198961.HKD":"SCHRODER ISF CHINA OPPORTUNITIES \"A\" (HKD) INC QV","LU0359201612.USD":"贝莱德中国基金A2","LU0244354667.USD":"SCHRODER ISF CHINA OPPORTUNITIES \"A\" ACC","LU1251922891.USD":"NINETY ONE GSF ALL CHINA EQUITY \"A\" (USD) ACC","LU0359202008.SGD":"Blackrock China Fund A2 SGD-H","LU0051755006.USD":"摩根大通中国A (dist)","01024":"快手-W","LU1720050803.USD":"安联全方位中国股票基金","LU1023057109.AUD":"BGF CHINA \"A2\" (AUDHDG) ACC","LU0210526637.USD":"JPM CHINA \"A\" (USD) ACC","LU0326950275.SGD":"Schroder ISF China Opportunities A Acc SGD-H","LU0456827905.SGD":"JPMorgan Funds - China A (acc) SGD","LU1303224171.USD":"NINETY ONE GSF ALL CHINA EQUITY \"A\" (USD) INC","BK1610":"ETF&股票定投概念","LU2097828557.USD":"AZ EQUITY CHINA \"A\" (USD) ACC","LU1719994722.HKD":"NINETY ONE GSF ALL CHINA EQUITY \"A\" (HKD) ACC","BK1608":"元宇宙概念"},"source_url":"http://www.zhitongcaijing.com/content/detail/1348592.html","is_english":false,"share_image_url":"https://static.laohu8.com/e9f99090a1c2ed51c021029395664489","article_id":"2569979252","content_text":"智通财经APP获悉，当前，电商平台普遍采用“召回、粗排、精排”的级联式搜索架构。该架构虽然成熟稳定，但仍面临诸多痛点：商品描述混乱、相关性问题突出、级联结构存在瓶颈以及冷启动难题，导致搜索结果往往不尽如人意。为了解决这些挑战，快手(01024)提出了业界首个工业级部署的电商搜索端到端生成式框架——OneSearch，目前，该系统已在快手多个电商搜索场景中成功部署，每日服务数百万用户，产生数千万页面浏览量。打破传统架构 提出创新解决方案OneSearch框架集三大创新于一身：关键词增强层次量化编码(KHQE)模块、多视角用户行为序列注入策略以及偏好感知奖励系统(PARS)。在关键词增强层次量化编码(KHQE)模块中，采用RQ-OPQ编码方案，从纵向与横向两个维度建模商品特征，为每个商品生成具备丰富语义层次的“智能身份证”，极大提升生成式检索的区分能力和准确性。多视角用户行为序列注入策略则让OneSearch能有效捕捉用户的近期偏好与长期兴趣，基于用户的长短期行为序列构建具有区分性的用户标识。使系统实现更全面、深层的用户意图理解，显著提升个性化搜索准确性与用户体验。偏好感知奖励系统(PARS)是结合多阶段监督微调与自适应奖励强化学习机制，捕捉细粒度用户偏好信号。该机制在提升排序性能的同时，保障生成多样性，有效避免“奖励破解”问题。订单量、买家数等多项指标实现大幅提升离线实验表明，OneSearch相比现有级联式系统各项指标均有显著提升。在线部署结果更为突出：订单量提升3.22%，买家数增长2.4%。这是在大规模工业场景下，这是生成式模型首次在大规模工业场景中完整替代传统搜索链路，具备重要落地意义。在人工评估中，OneSearch不仅在CVR和CTR上表现优异，还在页面整体满意度、商品质量及query-item相关性方面均显著优于传统级联系统。此外，系统在线性能同样亮眼：机器计算效率(MFU)相对提升达8倍，线上推理成本(OPEX)降低75.40%，资源利用效率得到极大优化。特别值得关注的是，OneSearch在冷启动场景下表现尤为突出，效果明显优于常规场景，表面生成式检索模型能够有效应对长尾用户和新上架商品的排序挑战。OneSearch的成功部署，标志着生成式模型在大规模工业搜索场景中取代传统链路的重大突破，为电商搜索技术的未来发展指明了方向。相关团队多项技术突破已发表在RecSys、CIKM、KDD等国际顶级会议上，引起业界广泛关注。后续，快手将持续探索在线实时编码方案，缩小预定义编码与流式训练之间的差异。同时，还将引入更强大的强化学习机制以更精准地匹配用户偏好，并结合图像、视频等多模态商品特征，进一步提升模型的推理效果与用户体验。随着技术的不断迭代，未来的电商搜索将更加智能、精准和“懂你”，让用户真正实现“一步到位”的理想搜索体验。","news_type":1,"symbols_score_info":{"81024":0.6,"01024":1.5,"KSTmain":1.5}},"isVote":1,"likeStatus":false,"favoriteStatus":false,"reportStatus":false,"tweetType":1,"langContent":"CN"}],"hasMoreComments":false,"newsSizeData":{"likeSize":1,"commentSize":1,"repostSize":1,"favoriteSize":0,"likeStatus":false,"favoriteStatus":false},"isCrawlerRequest":true}