[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"aihot-art-84421":3},{"itemId":4,"vertical":5,"category":6,"source":7,"score":8,"title":9,"summary":10,"analysis":11,"url":12,"coverUrl":13,"direction":13,"marketSignal":13,"publishedAt":14},"84421","ai","论文研究","Apple Machine Learning Research",60,"Apple Music搜索的多语言语义检索","Apple Music推出多语言语义检索系统，基于3.05亿参数双编码器模型，提升拼写错误和跨语言查询的召回率。","Apple Music 推出多语言语义检索系统，用 3 亿参数模型提升拼写错误、音译和跨语言查询的召回率。\n· 系统基于 3.05 亿参数的孪生双编码器，从 GTE-multilingual-base 微调而来，采用课程调度的多目标训练。\n· 针对 Apple Music 覆盖 150 多个国家、数十种语言、每日新增数十万首曲目的规模，重点优化长尾查询（占独特查询的大多数）。\n· 模型通过语义匹配而非关键词匹配，能理解“错误拼写”、“音译”和“跨语言”查询的意图，例如用户用英文搜中文歌名。\n· 集成到搜索栈后，显著提升了会话质量，尤其是对非母语用户和冷门内容。\n影响\u002F看点：该技术展示了多语言语义检索在音乐流媒体中的实用价值，为全球用户提供更自然的搜索体验，可能成为行业标准。","https:\u002F\u002Fmachinelearning.apple.com\u002Fresearch\u002Fmultilingual-semantic-retrieval",null,"2026-07-14 08:00:00"]