Recommender Systems and their Security Concerns

Instead of simply using two-dimensional User × Item features, advanced recommender systems rely on more additional dimensions (e.g. time, location, social network) in order to provide better recommendation services. In the first part of this paper, we will survey a variety of dimension features and show how they are integrated into the recommendation process. When the service providers collect more and more personal information, it brings great privacy concerns to the public. On another side, the service providers could also suffer from attacks launched by malicious users who want to bias the recommendations. In the second part of this paper, we will survey attacks from and against recommender service providers, and existing solutions.