TensorFlow Ranking框架在海外推荐业务中的实践与应用

{"type":"doc","content":[{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"前言"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在当今互联网世界，推荐系统在内容分发领域扮演着至关重要的角色。如何尽可能的提升推荐系统的推荐效果，是每个推荐算法同学工作的核心目标。在爱奇艺海外推荐业务，引入TensorFlow Ranking（TFR）框架，并在此基础上进行了研究和改进，显著提升了推荐效果。本文将分享TFR框架在海外推荐业务中的实践和应用。"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"01 算法的迭代：从传统CTR预估到LTR"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"长期以来，在推荐系统排序阶段广泛应用的CTR预估算法的研究重点在于，如何更加准确的估计一个用户对于一个item的点击概率。在这类算法中，我们将一组同时曝光在用户面前的items，当做一个一个单独的个例看待，将用户的特征、环境特征和一个一个item 的特征分别组合成为一条条训练数据，将用户对这个item的反馈（点击、未点击、播放时长等）作为训练数据的标签。这样看似合理的问题抽象其实并不能准确的表征推荐场景。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"严格来讲，排序问题的本质（尤其是以瀑布流形式呈现的业务）并不是研究估计一个用户对于一个单独的item的点击概率，而是研究在一组items同时曝光的情况下，用户对这组items中哪个的点击概率更大的问题。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}}]}