PriCoRec balances ad recommendation accuracy with privacy
New collaborative framework keeps sensitive user data on-device while maintaining 90%+ recommendation performance.
PriCoRec is a privacy-aware ads recommendation framework that keeps sensitive user features such as age and gender on-device, while still personalizing recommendations. It splits the work: a cloud-based pre-ranking stage uses cloud-accessible features, and an on-device ranking stage brings in highly personalized features locally. To improve candidate quality, the framework adds a diversity regularizer to pre-ranking, and it uses a cloud-guided training mechanism to keep the device model lightweight without sacrificing performance. According to the article, experiments show PriCoRec maintains strong recommendation performance while keeping sensitive features on-device.
- PriCoRec splits ad recommendation into cloud pre-ranking and on-device ranking to protect sensitive user data like age and gender
- Incorporates a diversity regularizer and cloud-guided training to maintain 90%+ recommendation accuracy with lightweight on-device models
- Accepted to RecSys'26; proposed by researchers from UCD, ByteDance, and 8 other institutions
Why It Matters
Enables privacy-compliant ad targeting that meets GDPR-like regulations without sacrificing recommendation quality or user personalization.