Research And Design On Hybrid Collaborative Filtering Algorithm And Its Attack Detection Model

Collaborative filtering as the key algorithm plays an important role in recommendation systems. It can not only help people to retrieve valuable resources from massive amounts of data, but also can provide enterprises the personalized recommendation services from a variety of businesses. However, there are still several deficiencies and drawbacks in the collaborative filtering algorithms in the aspect of accuracy and security.Firstly, we introduce a hybrid method which combines clustering and slope-one algorithms in order to improve the recommendation precision and quality. Our proposed hybrid algorithm uses item tags to reduce the sparsity of rating matrix and promote the recommendation rate. In addition, the experiment conducted by the data set from MovieLens shows that the hybrid method can improve the accuracy of recommendation system.Secondly, for security concerns, shilling attack is the most serious threat in recommendation systems. Due to the openness of recommendation system, it is vulnerable to injected fake information and affects its accuracy and reliability. In order to fix the security bug, this thesis also introduces the ant colony optimization to construct an anti-attack model. By using ant colony algorithm, this thesis also realizes a trust model which could effectively prevent the shilling attack and enhance the security guarantee of the recommendation system. Finally, the experiments and analysis on our system prototype based on hybrid CF and anti-attack model show that our proposed algorithm and model can provide more user-friendly and comfortable experiences for users, and support more precise and perfect recommendation services for enterprises.