Distributed Collaborative Filtering Protocol Based on Quasi-homomorphic Similarity

We study the problem of predicting the rating for an unseen item based on distributed dataset by two honest-but-curious parties without revealing each private dataset. Our proposed idea uses a new similarity measure such that similarity aggregated with two local similarities is approximately equal to the global similarity. We show the accuracy reduction and the performance gain given by our proposed scheme based on an experimental implementation, and claim that our scheme allows parties to estimate prediction in a practical model with negligible accuracy reduction.