A fuzzy variant of k-member clustering for collaborative filtering with data anonymization

Privacy preserving data mining is a promising approach for encouraging users to exploit the IT supports without fear of information leaks. k-member clustering is a basic technique for achieving k-anonymization, in which data samples are summarized so that any sample is indistinguishable from at least k - 1 other samples. This paper proposes a fuzzy variant of k-member clustering with the goal of improving the quality of data summarization with k-anonymity. Each k-member cluster is extracted considering the fuzzy membership degrees of samples, which are estimated based on the distance from clusters. The proposed anonymization method is also applied to collaborative filtering, in which the main task is estimation of the applicability of unevaluated items. Several experimental results demonstrate the characteristic features of the proposed anonymization method.