A privacy-preserving collaborative filtering protocol considering updates

This paper proposes a method to update the similarity of items in a privacy preserving collaborative filtering. The similarity of items is a value that shows how similar given two items are. Privacy preserving collaborative filtering is a technique that helps to infer an evaluation value of desired items given the other users´ evaluation values with concealing personal information for each user´s privacy by encrypting the evaluation values. In order to obtain the most appropriate evaluation value, it is necessary to update the similarity every time when an evaluation value is changed. Since each evaluation value is encrypted, it is a heavy burden for the users to update the similarity of items every time in response to a single change of evaluation values. Thus we need to know when we should recalculate the similarity of items while keeping each renewal of evaluation value secret. In this paper, we show that the estimation error of an evaluation value is small if the error of the average of evaluation values between the users is small. We show an algorithm that detects a change of the average of the evaluation value that is greater than the preset, using a constant number of plaintext equality tests for each renewal of evaluation values.