A Collaborative Filtering Recommendation Method Satisfied with Differential Privacy

Abstract

The invention discloses a collaborative filtering recommendation method satisfying differential privacy. The method comprises the steps of firstly, calculating a grouping probability of a preference project of a target user, and according to the value of the grouping probability, determining whether the project is classified into a random set or a neighbor set; secondly, according to a replacement probability P', determining to execute three processes including arbitrary project replacement, neighbor project replacement and existing project retaining separately to complete the reconstruction of user's preference, and choosing neighbors of the target user according to a reconstructed preference application index mechanism; finally, generating a recommendation result according to the neighbors. The collaborative filtering recommendation method satisfying the differential privacy can ensure that the privacy of the user is not disclosed in the best way.

Description

Technical field [0001] The invention relates to the field of privacy security, in particular to a collaborative filtering recommendation method that satisfies differential privacy. Background technique [0002] According to the KNN attack model, it is known that the attacker modifies the historical record information twice to speculate whether the recommendation result exists in the historical item set of the attacked user. Even without considering the accuracy, after simply applying differential privacy protection in collaborative filtering recommendation, any attacker or malicious user cannot know whether a particular user has a score record for an item. However, an attacker or malicious user is very likely to use the recommendation system to infer whether an item is similar to an item with a scoring record. Therefore, the attacker or malicious user can infer more accurate user preference information without knowing exactly which items the user has evaluated. Therefore, in th...

AI Technical Summary

Problems solved by technology

[0004] In the current privacy-preserving recommendation methods, there are problems that only consider the unilateral enhancement of privacy protection effect while ignoring the recommendation accuracy, and the problem that the recommendation accuracy is guaranteed but the privacy protection is not strong enough.

Images