Search papers, labs, and topics across Lattice.
This paper tackles the complex challenges of recommendation systems in live-streaming by introducing a multi-objective ranking approach that balances fresh and delayed user signals. The authors implement a delayed window strategy for feedback collection, a multi-model architecture that integrates various user behaviors, and a segment-aware targeting module that adapts ranking scores based on user lifecycle stages. Their online A/B testing reveals significant performance gains, including a 0.09% increase in Daily Active Viewers and a 1.12% rise in positive interactions on the Twitch mobile feed, showcasing the method's effectiveness across diverse user segments.
By integrating fresh and delayed signals, this approach boosts viewer engagement and revenue while reducing model complexity by nearly 42%.
One of the most challenging problems entertainment live-streaming services face in recommendation systems is that user behaviors are sparse and delayed, and interaction data exhibits bias for different user segments. Unlike e-commerce applications where user actions follow linear sequences, live-streaming viewers engage in multiple concurrent behaviors of watching, chatting, following, and spending, each occurring with varying delays. We address these challenges through three key contributions: 1) a delayed window approach that extends feedback collection beyond immediate responses, 2) a multi-model architecture that combines fresh and delayed signals, and a segment-aware targeting module that optimizes ranking scores differently across user lifecycle stages, and 3) Multi-gate Mixture-of-Experts (MMoE) integration that jointly models correlated targets while reducing model parameters by 41.9% compared to independent models. Online A/B testing demonstrates significant improvements, including a +0.09% increase in Daily Active Viewers (DAV), generating millions more annual active viewer days, and +0.56% increase in highly engaged viewers'capped Average Revenue Per User (ARPU). Viewer-segment targeting achieved an additional +0.15% DAV improvement for newer and less engaged viewers, while MMoE enhancement added +0.08% overall DAV and +0.27% new follows. The proposed system processes ranking requests with low latency, providing a scalable approach for balancing multiple business objectives across diverse user populations. In addition, we tested the multi-model architecture on the Twitch mobile live feed and achieved a +1.12% increase in positive user-channel interactions (clicks, follows, and likes), demonstrating applicability beyond the primary use case.
Amazon Science