Study Of Cross Recommendation Based On User’’s Dynamical Interest And Variable Weight Privacy-preserving Algorithm

In this era of information explosion, both users and information producers of media are facingthe new challenges.To address this issue, the personalized search and information recommendation technologyare introduced by the major e-commerce platforms. It can help users to get attention information, which solve the problem of information overload of the Internet, to provide users with product conforms to their interests and preference, which according to user’s preset parameters or the records to obtaineduser’s interests and preferences by data mining. Ideally, this recommendation system can greatly reduce the user’s search costs, as well as, provide users with personalized service, so as to improve the user’s satisfaction for shopping.However, the traditional recommendation system cannot achieve the desiredeffect of products recommendationsaccording to the user’s historical information and profile. Thetraditional recommendation system that is based on collaborative filtering algorithm only considers the similarity of user’ interest without consideration of the dynamic interests, as well as, only considers the users’ preference of goods without considering the differences preference of users’ search behavior. At the same time, depth data mining in across the e-commerce platform brings great hidden trouble to user’s privacy.Therefore, this thesis tries to explorerecommendation with the dynamic interest and variable weight privacy-preserving algorithmto achieve the cross recommendation in the cross-platform based on the theoreticalfoundation of identify the dynamic interest and the complex behavior. Specifically, this thesis tries to estimate the user behavior preferences and to identify the user’s dynamic interest from the user’s online behavior trajectories and consider variable weight security multi-party computationprivacy protection in the process.In order to effectively improve the existing recommendation system that only considers the steady-state user preferences and transient behavior preference and to reduce the high cost of data collection, to make the recommendation results more in line with the user’s current personal intention. The specificcontents have been show as follow,First of all, this thesis embarks from the dynamic interest to build the timing interest graph to provide higher quality of personalized service. It estimates the user’s interest degree based on the method of acquiring user’s browsing behavior. So the browsing behavior is estimated on the basis of user interest degree, and points out that the indirect behavior is main source of behavior. Through the correlation analysis of indirect behavior find clustering of indirect behavior.Then propose the estimation method based on the scale of residence time and interest based on browsing speed.Secondly, the server to collect users’ information or personalized recommendation led to a serious privacy problem, this paper put forward variable weight secure multi-party privacy-preserving algorithm, to ensure the mining quality ofuser’s interested, at the same time it can protect the user’s personal sensitive information. Specifically, based on the acquisition and analysis of dynamic user interest, this thesis considers behaviormodeling and classification of user privacy preference. Then, I put the weighting function into the privacy of secure multi-party computation, to ensure that the user’s private information is protected and not be infringed.Finally, this thesis integrates the user’s interest based on cross-platform, to improve the quality of e-commerce service on personalized recommendation.Explore formation mechanism that shares between each platform and the dynamic model of the different types of users’ final purchase decision, in order to better implement the cross recommendation in cross-platform.The innovative contributions of this thesis have been concluded as follow,· The user’s dynamic interest and interest changing law are introduced into the recommendation system.· Considering users’ dynamic privacy problem in data mining, and proposes the variable weight privacy-preserving algorithm based on secure multi-party computation.· Set up the across-platform recommended model based on complex network privacy protection to achieve the high satisfaction and increasing service value of electronic commerce and contributions foundation of theory and practice.