A privacy-preserving collaborative filtering approach using Hellinger distance similarity metric for high dimensional dataset

By proposing items or particulars to users, collaborative recommenders (CR) have become significant on internet applications. Recommender algorithms suggest recommendations to users, that are utmost similar to those earlier consumptions. To suggest recommendations, the system requires users' sensitive information such as browsing history, purchase activities, and demographics. Unfortunately, this information may be misused by malicious third parties for endorsing their products. In this paper, we present a privacy-preserving recommender system approach to protect sensitive information against breaches. Our framework is initiated on homomorphic encryption, which is used to conceal the sensitive information of the users from third parties. While the user's preferences, favourites, like-tasted neighbourhood, and suggestions are transmitted over the network in an encryption form. However, traditional neighbourhood CR uses Jaccard similarity faces difficulty in handling high-dimensional and sparse data. The proposed approach diminishes these issues by exploiting the Hellinger similarity metric. The efficacy of the proposed model is accurate as regards computation also outperforming existing traditional solutions.