A Study on Privacy-Preserving Transformer Model for Cross-Domain Recommendation

As customer relationship management becomes increasingly data-driven, cross-domain recommendation (CDR) systems are critical in leveraging insights from different domains to enhance customer experience. However, the aggregation of cross-domain user data raises significant privacy concerns. We propose a transformer-based CDR model that shows improved performance on key metrics such as Mean Reciprocal Ranking (MRR) and hit rate. In this model, we introduce a privacy-preserving method using embedded mask and differential privacy to protect user information. Our contribution is twofold: we propose ways to protect user privacy in CDR and analyze the balance between accuracy and privacy.