Researchers have developed a novel multi-objective ranking system for live-streaming recommendation platforms to address challenges of sparse and delayed user behaviors. The system incorporates a delayed window approach for extended feedback collection and a multi-model architecture that combines fresh and delayed signals. Additionally, a segment-aware targeting module optimizes ranking scores based on user lifecycle stages, and a Multi-gate Mixture-of-Experts (MMoE) integration reduces model parameters while jointly modeling correlated targets. Online A/B testing showed significant improvements in Daily Active Viewers (DAV) and Average Revenue Per User (ARPU), with specific gains for newer viewers and overall DAV increases from the MMoE enhancement. AI
IMPACT This research could improve user engagement and monetization for live-streaming platforms by optimizing recommendation algorithms.
RANK_REASON The cluster contains a research paper detailing a new system for live-streaming recommendation. [lever_c_demoted from research: ic=1 ai=1.0]
Read on arXiv cs.IR (Information Retrieval) →
- arXiv
- Multi-gate Mixture-of-Experts (MMoE)
- Multi-Objective Ranking for Live-Streaming: Balancing Fresh and Delayed Signals with Segment-Aware Targeting
- Twitch
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
