Rohan Paul (@rohanpaul_ai) · 2026/10/20 13:45:16
X 调整推荐算法 强化停留时长与完读率
X 平台对其“为你推荐”信息流进行了两项关键算法调整，旨在最大化用户停留时间。新的排序模型更精准地预测潜在观看时长，而检索模型则转向优先展示用户倾向于完整阅读的内容。这一变化导致非关注账号内容的筛选逻辑重构，并伴随平台日活跃秒数显著增长至约5000亿。
报道全文原始报道全文
Rohan Paul (@rohanpaul_ai) · 2026-10-20 05
X has changed 2 stages of its For You feed to favor time spent and posts people finish.
-
Musk said the new ranking model “predicts probable time spent much better”
-
Dwell time was already among the ranker’s predicted actions before this update, which means the new model sharpens an existing objective.
-
The second change acts earlier, at retrieval, which turns you and each post into vectors and pulls the nearest ones from accounts you don’t follow. Posts from accounts you follow arrive through a separate system, so this retrieval model mainly decides which posts from accounts you don’t follow make the shortlist.
-
Musk said this step is now trained on read-through, so it learns to pick posts that people tend to read to the end. A post that people usually abandon halfway is therefore less likely to make the shortlist.
-
daily user active seconds is seeing quite a massive growth, 2026 climbs from roughly 410B a day in mid-August to about 500B by mid-October,
↳ 引用 @elonmusk (Elon Musk): Two important changes to timeline recommendations:
New AI/SI ranking model that predicts probable time spent much better
Retrieval model trained to surface posts people read all the way through
User active seconds is growing exponentially and hitting all-time highs almost every day:
