An LSH-based Privacy Preserving Personalization Approach
06 February 2012
We present a solution to the "Privacy versus Personalization" dilemma, that enables the end-user to avail of collaborative personalization services without disclosing sensitive information to the content/service provider or any third party for that matter. Solving this dilemma is challenging primarily because generating collaborative-filtering recommendations requires the global profiles of all end-users in order to identify similar users, and to compute the top-rated items of the community of like-minded users. In this work, we observe that the Locality Sensitive Hashing (LSH) technique of scalably finding nearest-neighbors can be adapted in a novel way to enable discovering similar users in a privacy preserving way. We present extensive performance evaluations of an LSH-based privacy preserving recommender system on the basis of MovieLens dataset, and we show the trade-off that exists between the privacy protection level and the recommendation quality.
