Personalized Recommendations Scheme Based On Privacy Protection In Social Networks

The rapid development of mobile Internet technology has brought great convenience for people to obtain all kinds of information.With the explosive growth of users' exposure to information,personalized recommendation services were born.It can provide users with matching items based on their personal and behavioral information.Reasonable recommendation service can not only be liked by users,but also promote the development of the industry.The full use of user information is the basis of personalized recommendation services,and the user's behavior and attribute information are also the private information that inconvenient to disclose.In recent years,there have been incidents in which attackers deliberately steal user information in social networks and use user characteristic information through statistical analysis to pose a threat to user property and information security.Therefore,how to fully protect user privacy while achieving accurate recommendations has become a new research topic.When new users access the platform,the system lacks a large number of data attributes of the new users.How to quickly recommend content that satisfies the new users needs to be further studied.In response to the above-mentioned problems,this article conducts research from the aspects of recommendation quality improvement,user privacy protection,and user cold start.The main contents and innovations are as follows:(1)A collaborative filtering recommendation based on covariance clustering(COP-CCF)is proposed for providing efficient and diversified personalized services for users in social networks.COP algorithm is proposed for offline clustering,which divides users into specific clusters.Then,the vector cosine method is used to calculate the similarity between the offline basic users and each cluster center,and the interest agreement of offline basic users Variance measurement matrix.Aiming at online target user recommendation,CCF algorithm is proposed to achieve target user score prediction and obtain recommendation results.Simulation experiments show that the scheme can fully excavate user interest under the condition of sparse data and improve recommendation accuracy.(2)Aiming at the privacy protection requirements of recommendation systems in social networks,an initial point-optimized differential privacy protection scheme based on reachable distance clustering ranking(OIC-DPOP)is proposed.The scheme proposes the OIC-DPOP differential privacy protection algorithm to process the data.First,the initial point of the clustering algorithm is optimized,and the cluster analysis is performed by generating an extended cluster order,which neither explicitly generates data clustering results nor Output the parameter settings under a specific situation to effectively prevent attackers from stealing specific personal privacy information.Simulation experiments show that user data availability and privacy protection effects are effectively balanced and algorithm efficiency is improved.(3)To address the recommendation problem of cold start for initial target users in social networks,a cold start recommendation scheme based on privacy histogram publishing(ST-HP)is proposed based on the OIC-DPOP scheme.According to the clustering results of OIC-DPOP,the histogram statistics of each cluster are carried out,and the ST-HP algorithm is proposed.According to the histogram of each cluster and the set threshold,the neighbor clusters of the cold start user are selected,thereby calculating the nearest neighbors of the cold start user to generate recommended results.Simulation experiments show that the accuracy and rate of recommendation of this scheme have been greatly improved.At the same time,the change of the threshold can adjust the recommendation rate,recommendation breadth and recommendation accuracy,thereby increasing the flexibility of the algorithm and better solving the cold start problem of the initial user.