Differential privacy protection method for collusion inference attacks in collaborative filtering

Abstract

The invention discloses a differential privacy protection method for collusion inference attacks in collaborative filtering, which comprises the following steps of: grouping users according to user similarity, recommending the users according to a grouping result, associating strongly similar users into an integral user on the basis, and returning the same recommendation result to the integral user; making sensitivity calculation before differential privacy noise is added, and analyzing the possible maximum influence of user data on any other data recommendation result; adding differential privacy noise to the recommendation functionality, and carrying out recommendation for the user according to the interference functionality. Users who may be subjected to collusion attacks are associatedand subjected to unified interference, inference attacks are prevented, the differential privacy noise is reduced on the basis, and the recommendation accuracy is improved.

Description

technical field [0001] The invention relates to a differential privacy protection method, in particular to a differential privacy protection method against collusion inference attacks in collaborative filtering. Background technique [0002] In recent years, with the rapid development of the Internet, network information resources have grown exponentially, making it difficult for users to effectively obtain useful data from massive resources. Collaborative filtering uses the preferences of groups with similar interests and experiences to recommend content that users are interested in, and has become the most commonly used recommendation technology at present. However, while collaborative filtering helps users discover potentially interesting content from massive information resources, it also brings privacy leakage risks to users. Since each user’s recommendation results come from users with similar preferences, a malicious user can create some fake accounts similar to the ...

AI Technical Summary

Problems solved by technology

[0005] Therefore, when defending against user privacy leaks caused by collusion inference attacks, existing work protects users on the premise that users are independent from each other, either adding more noise data or randomizing users into groups. Will seriously affect the recommendation accuracy

Images