Paper 2024/106

A Trust-based Recommender System over Arbitrarily Partitioned Data with Privacy

Abstract

Recommender systems are effective mechanisms for recommendations about what to watch, read, or taste based on user ratings about experienced products or services. To achieve higher quality recommendations, e-commerce parties may prefer to collaborate over partitioned data. Due to privacy issues, they might hesitate to work in pairs and some solutions motivate them to collaborate. This study examines how to estimate trust-based predictions on arbitrarily partitioned data in which two parties have ratings for similar sets of customers and items. A privacy- preserving scheme is proposed, and it is justified that it efficiently offers trust-based predictions on partitioned data while preserving privacy.

Note: The highlights of our paper is as follows. 1- First of all, we investigate trust concept in terms of artificial intelligence, e-commerce and recommender systems and give insights in this context. 2- We showed how to provide trust-based recommendations on arbitrarily partitioned data between two online vendors with privacy. 3- A privacy-preserving scheme was proposed and theoretically justified that it efficiently offers trust-based predictions on partitioned data with privacy. 4- Empirical outcomes showed that collaboration improves both accuracy and coverage and the proposed scheme helps data holders provide more accurate referrals than the ones provided from split data only.