Multi-Party Secure Collaborative Filtering for Recommendation Generation
Harmanjeet Kaur, Neeraj Kumar, Mohammad S. Obaidat · 2019
Recommender systems based on collaborative filter- ing technique generate the accurate and reliable predictions for customers by using their preferences about various products. Usage of user ratings has been a prevalent and successful practice for various e-commerce based websites, but launching a new e-commerce site will not benefit from this, as new site has no database of customers. A combined effort by existing companies and newly launched companies for recommendation generation can be beneficial for both companies and customers, if confidential data is protected. In literature, most of the existing techniques for secure prediction generation are based on data distortion and homomorphic encryption techniques, which may cause accuracy loss and high computation cost, respectively. To overcome these issues, this paper proposes a prediction generation scheme for the horizontally distributed data among different companies, which can help new entrants and existing sites, while preserving the privacy of the customer data. Analysis of the proposed scheme is done for parameters: privacy, accuracy, coverage and performance. The proposed scheme is secure, and the accuracy and coverage are improved due to collaboration of multiple parties. Moreover, the computation complexity of the proposed scheme is also low.