The Research Of User-oriented Privacy-Preserving Collaborative Filtering

With the rapid development of economy,science and technology,Internet technology has entered into a new era.Recommendation technology appears in this context.Collaborative filtering technology is one of the most widely used and successful recommendation techniques for recommendation systems.Collaborative filtering based recommendation system must get a lot of user historical data to obtain accurate recommendation results.This process directly or indirectly leads to the risk of exposure of users' privacy.Collaborative filtering has been widely used in recommendation for a single user.In recent years,it has also been applied to group recommendations for group users.Group members have also shared privacy issues while sharing and interacting with each other.So user privacy protection should be taken into consideration in group recommendation.To address the issues above,this thesis makes a detailed research on the user-oriented privacy-preserving collaborative filtering.Main works are as follows:First of all,the thesis puts forward the random perturbation technique addressing privacy requirements for collaborative filtering.Users are classified into different groups according to users' privacy requirements.Users can choose the appropriate strength of noise according to their own privacy requirements.This reflects users' control for private data and users' difference for privacy requirements.On this basis,the thesis puts forward a method of weight setting based on time characteristic of privacy.Therefore,disturbance of noise no longer has constant distribution characteristics.The risk of disclosure of users' privacy information has further reduced.Further,the random perturbation method is applied into collaborative filtering system.By modifying formula of collaborative filtering,accuracy of prediction is improved.Simulation experiments and the results show that the proposed approach can protect user privacy data and at the same time obtain better recommendation results.Secondly,in order to solve privacy protection problem in the group recommendation,this thesis tries to apply the proposed random perturbation technique into collaborative filtering-based group recommendation and puts forward a novel collaborative filtering-based group recommendation method by employing random perturbation technique.To protect users' privacy information,the thesis ensconces users' original data by random perturbation technique and groups via the users' interests and preferences similarity.The group preference is obtained by the weight of the users' preferences and the weight of the frequency.Predicted group rating is obtained by collaborative filtering based on item.The proposed method protects users' privacy information in all the groups and at the same time achieves group recommendation.Recommendation accuracy and efficiency has been improved compared to other traditional methods.Finally,this thesis constructs a privacy protection collaborative filtering prototype and designs an application demonstration based on the above theory and technique.Application demonstration shows the demand analysis,conceptual design,detailed design and implementation process.Privacy type management,disturbance weight management,and other function modules have been designed.Feasibility of the proposed method has also been verified.The prototype shows that user-oriented privacy-preserving collaborative filtering is effective in practical applications.