最近但非最爱：面向纯内容搜索与推荐的共享策展反馈基础设施
原标题：Nearest but Not Dearest: Shared Curator-Feedback Infrastructure for Content-Only Search and Recommendation
AI 摘要
Feed.fm 的 B2B 音乐发现平台在无听众行为信号的内容检索场景下，发现余弦最近邻有 38% 被策展人拒绝。作者将拒绝原因分为声音失败（55%，风格/节奏/情绪不匹配）和语境失败（37%，语言/节日/版权等），并分别路由到候选生成层的约束过滤器和表示层的嵌入重加权头。在 1200 条策展判断上，该干预将拒绝率从 38.17% 降至 28.83%（相对下降 24.5%），但作者强调这是生产环境下的会计边界而非严格因果估计。
正文节选
Nearest but Not Dearest: Shared Curator-Feedback Infrastructure for Content-Only Search and Recommendation Abstract. A deployed B2B music-discovery platform serves both query-driven search (semantic-text prompts, vibe-tag selection) and seed-driven recommendation (seed-track and artist stations) over one licensed catalog, one LAION-CLAP joint audio–text embedding space, one candidate-generation filter, and one ranking head — and neither path consumes end-listener behavioral signal. In this conte