CAPTS: Channel-Aware, Preference-Aligned Trigger Selection for Multi-Channel Item-to-Item Retrieval
This new framework is fixing a major flaw in how TikTok and YouTube recommend videos.
Deep Dive
Researchers from Kuaishou (Kwai) have published CAPTS, a new AI framework that optimizes multi-channel recommendation systems. It solves two key problems: biased value attribution and uncoordinated routing between channels. In large-scale online A/B tests on Kwai's short-video platform, the system delivered a significant +0.351% lift in average watch time per device. The framework treats trigger selection as a learnable routing problem to maximize downstream engagement.
Why It Matters
This directly improves user engagement and revenue for major video platforms, making recommendations more efficient and personalized.