Privacy Preserving Collaborative Filtering Approach for Recommendation System

Abstract

Recommender Systems use collaborative filtering in order to make recommendations based on similar interest between users or items. In this process, Privacy of users is at severe risk because recommender servers may share user’s private data with third parties to make personalized advertisements and in some cases user privacy may be exposed to the public or attacked by malicious users. The existing works are based on encryption-based and randomization-based techniques, but they compromise accuracy for privacy and privacy for accuracy. In this project, Privacy Preserving Collaborative Filtering approach has been proposed which solves the limitations in the existing works. This work adopts fuzzy logic to deal with uncertainty among user’s interest ratings. The fuzzified data is then perturbated by applying random rotation perturbation technique, thus the user’s interest is not directly available for the recommendation server or the third-party. Using the perturbated data, item clusters are formed by utilizing ant-based clustering algorithm. These clusters help the recommendation server to apply item-collaborative filtering for the recommendation process. In order to refine the cluster center provided by ant-based clustering process K-Means clustering algorithm is applied. The pheromone values obtained during the ant-based clustering is further utilized by the recommender server in order to provide accurate recommendation to the active user.