Earticle

Personalized recommendation algorithm is core to recommendation systems, which matters the quality of recommendations of such system. The paper proposed an improved Slope one recommendation algorithm, M-Slope one. Based on real-time user interest model, the new method can calculate similarities between users and establish neighboring user groups to narrow down search scope of related items and improve the average rating differential equation for items. The algorithm proves its effectiveness for improving the precision of recommendations.

In view of the high vulnerability of traditional user-based recommendation algorithm to shilling attacks, In this paper, on the basis of the work of the group effect on the attack profiles, this paper analyzes the statistical features of the nearest neighbors of target users before and after attack, Design a kind of Attack Profiles online filter to attack the target user profile from the nearest neighbor filter. And this filter improves the user-based recommendation algorithm nearest neighbor selection strategy, thus proposes the Collaborative Recommendation algorithm based on Online Filter for Attack Profiles (CROFAP). Experiments show that attack profile online filter can accurately identify and filter out most attacks profile to ensure the robustness of the CROFAP algorithm.