Research On Privacy Protection In Mobile Application Recommender System

Wide spread of smart phones accelerates the quick development and a broad range of requirements of mobile applications.However,there are too many mobile applications in App stores,and applications with same or similar functionalities vary from types and qualities,which results in information overload in App market.Mobile users are confused in choosing suitable and trustworthy mobile applications due to a large number of available candidates.A mobile application recommender system is a system that applies the technology of recommender into mobile applications and is a powerful tool that helps users solve the problem of information overload.However,there are few feasible recommender systems focusing on recommending mobile applications in the literature.First,few researches study user trust behavior based recommendation on mobile applications.User trust behavior of using mobile applications can greatly imply user preferences,and thus the data of user trust behavior are of great value when building user profiles and generating recommendations.Second,the accuracy and personalization of existing recommender systems need to be further improved.The existing basic rule in App stores is based on the number of download and user ratings,the accuracy of which,however,is low while the risk of hostile attacks exists.Particularly and importantly,privacy preservation is still an open issue in mobile application recommendation.Recommendations are generated based on the computations on user data,even private personal data,so privacy leakage could occur without effective data protection.Aiming at solving the problems mentioned above,in this paper,we propose two privacy-preserving mobile application recommendation schemes based on trust evaluation.The proposed two schemes are able to preserve user privacy data while providing mobile application recommendations,thus helping mobile users solve the difficulty of choosing trustworthy mobile applications.Different system structures and working procedures are designed for the two schemes,in order to apply them into different scenarios.The first scheme has a centralized system structure with three types of system entities and could be used in situations based on cloud services.And the second scheme fits into a distributed system structure that consists of two types of system entities,which could be used in distributed scenarios such as social networking.Recommendations on mobile application are generated based on user trust behaviors of mobile application usage,in which the accuracy and personalization on recommendations can be greatly improved and the impact of hostile attacks can be also avoided.In the procedures of generating recommendations in these two schemes,user private data can be preserved by applying our proposed security protocols and utilizing homomorphic encryption,thus avoiding privacy leakage.We further implement two schemes and develop two mobile Apps that can be applied in different scenarios.What's more,security analysis referring to our security model demonstrates the security holding by our schemes.Besides,we evaluate the performance of two Apps developed based on the designed schemes regarding their efficiency,memory cost,CPU usage,communication cost,and battery consumption.Both of two Apps show better performance compared to existing related work.Finally,we conduct simulation tests where different proportion of attackers are simulated considering three common types of attack,i.e.,bad-mouth attack,on-off attack,and conflict behavior attack.The simulation result shows that our schemes can resist above attacks at a good level and have sound robustness.