Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·Aug 18·5 min read
Decomposing Staleness in Recommender Systems: A Dual-Filter Framework for Supersession and Decay
TL;DR
A new filtering system called SDF (Supersession-Decay Filtering) was developed to tackle stale recommendations in Google Discover, significantly reducing user complaints. Over two years, user-reported staleness dropped by 54.9%, enhancing user engagement.
✦ Why It Matters
Implement SDF-like filtering in your recommendation systems to enhance user satisfaction and reduce complaints.
Key Takeaways
- Implement dual-filter systems to address content staleness.
- Use relational models to detect supersession in recommendations.
- Train models on historical traffic data for better relevance predictions.
- Monitor user feedback to continuously improve recommendation quality.
How It Works
SDF employs two models: the relational staleness model identifies when new content makes older items irrelevant, while the predicted traffic ratio (PTR) model forecasts how an item's relevance decays over time based on its historical engagement. By filtering out stale items before ranking, SDF ensures that users receive more relevant recommendations.