A Study On The Utility Of Recommender Systems That Weigh Privacy

Nowadays,with the continuous development of the Internet and information technology,the Internet has become an indispensable part of people's life,infiltrating into all aspects of life and work.People have produced hundreds of millions of information through various ways.In view of such a large amount of information,it is difficult for users to find useful information quickly,which thus leads to the information overload.Recommendation system can solve information overload,because recommendation system is based on massive personal information and behavior data of users,using statistical analysis,machine learning and deep learning algorithms to analyze and predict users' preferences,and recommend potential relevant items to users.The recommendation system can reduce the search cost of users to a large extent,and make people quickly locate the useful information for themselves,so as to bring convenience to users.However,recommendation system is a double-edged sword,that is,it brings people great convenience,but also has some problems.It needs to collect a large number of user information,which may have the risk of leakage,and the recommendation system may not collect user information in the case of explicitly informing users.When users use the platform or recommendation system,they will feel that their personal information is collected and used by the platform,so as to reduce the effectiveness of the recommendation system,and then take actions such as reducing the number of times of using the recommendation system,unwilling to share personal information,providing false data,etc.These behaviors will have a great impact on the accuracy of the platform and recommendation system,which violates the original intention of designing recommendation system and cannot improve the platform revenue.Therefore,this paper wants to explore the impact of collecting user information on user utility and platform revenue by also taking privacy risk into consideration.In this paper,in a recommendation system,we aim to study and analyze how and what kind of user information should a platform to collect and thus can maximize its revenue.First of all,from the influence of user information on recommendation accuracy and privacy concern,the mathematical model of recommendation accuracy and privacy concern is established by taking user information as independent variables.Then,based on the utility theory,according to the influence of recommendation accuracy and privacy concern on user utility,the mathematical model of the influence of two intermediate variables,i.e.,recommendation accuracy and privacy concern,on user utility is constructed.Then,according to the relationship between user utility,recommendation accuracy and platform revenue,the mathematical model of target variable is established.Finally,we analyze the model via simulations towards the established model.Simulation results show that the combination of different user characteristics collected and used by the platform will have different effects on user utility and platform revenue,and there exists a feature combination to maximize platform revenue considering privacy risk.Results also reveal that there is an optimal feature number,which makes the average platform revenue reach the maximum value.Then,we extend the basic model by also considering the influence of other relevant factors,including privacy protection,privacy group category,correlation between features and time factor.Simulation results on extended models verify that all the four aforementioned factors will have an impact on user utility and platform revenue.