In the age of information explosive growth,science and technology has developed so fast.However,this growth rate is far beyond the scope of users and systems can accept,process and use effectively.And this is called information overload.With the ability of providing users with better interactive experience,recommender system is becoming more and more irreplaceable in online web services.And since it is established on users’ private information,the more abundant data it accumulates the higher accuracy it could achieve.However,recommender system also has to face some threats including users’ private data disclosure and predicament on their states and behaviors based on big data.Users’ interest change over time,so there is a problem of interest drift.And recommender effect will have a great impact,if we can not track change in users’ interest effectively.In order to protect the privacy and security,this paper introduces differential privacy theory,time series theory and constructs the recommender model.The main content of this paper is described as the following:(1)In order to solve the privacy and security problem,we introduce differential privacy theory.Differential privacy can solve two shortcomings of the traditional privacy model: Firstly,define a strict anti-attack model,regardless of how much background knowledge the attacker has.Assuming that the attacker has obtained all records except the target record,i.e.maximum background knowledge,the privacy of the record will not be disclosed.Secondly,it gives privacy protection a rigorous definition and provides a quantitative assessment method.We use differential privacy theory to add noise on data set,which satisfies the Laplace distribution.And it increases the privacy and security without affecting the recommended effect.(2)With the purpose of resolving the interest drift problem,we introduce time series theory.Recommender system is based on users’ historical data.Firstly,we analyze the relationship between users’ rating time and the earliest rating time and the latest rating time.Then construct the time function with rating time,depict the change of the users’ interest and increase the accuracy of the recommender model.(3)For the sake of the sparsity problem of data set,we fill up the matrix.As user-item matrix often has high dimensions,users are generally unable to rate most items,on the other hand each item can not be rated by most of users.Therefore matrix always has many missing ratings,then we fill the missing data with the average of the user-item matrix.(4)With the aim of solving the problem of users’ and items’ preferences,we introduce user and item biases.By adding user biases and item biases,we build preference model.Finally,we build recommender algorithm that combines the above four factors.Through the comparison of several groups of experiments,fully demonstrate that our proposed method has validity and higher accuracy,which provides a valuable perspective for privacy-preserving recommender research. |