Performance Comparison of Collaborative Filtering with k-Anonymized Data by Fuzzy k-Member Clustering

Abstract

In order to perform collaborative filtering with published databases in a privacy preserving manner, databases must be anonymized beforehand. This paper studies the applicability of fuzzy k-member clustering in privacy preserving collaborative filtering with k-anonymized data, in which users’ historical data of k or more users are suppressed considering soft data partitions. By allowing boundary samples to be shared by multiple clusters, data anonymization is performed without significant loss of information. Its performances are compared with several different types of fuzzy membership functions.