Personalized Recommendation Based On User Portrait And Privacy Protection

Personalized recommendation system is an indispensable part of modern e-commerce,it can recommend personalized goods and services for users according to their interests,preferences,historical behavior and other information.However,while providing personalized recommendation for users,the recommendation system also involves the privacy protection of users.How to provide better personalized recommendation service on the premise of protecting user privacy has always been a research hotspot in the field of recommendation system.In order to solve this problem,this thesis studies SVD++matrix decomposition in collaborative filtering algorithm,and introduces label vector and Gaussian mechanism into matrix decomposition model.The label vector represents the feature vector of commodity or user,which is matrix decomposed together with the implied features of user and commodity,so as to obtain the user’s score prediction of commodity.At the same time,in order to protect user privacy information,differential privacy Gaussian noise and SGD gradient perturbation processing are used in this thesis to prevent attackers from inferring attacks on model parameters by micro-perturbation of model parameters.The algorithm proposed in this thesis can improve the effect of personalized recommendation system on the premise of protecting user privacy.Experimental results show that the proposed algorithm can effectively protect users’ privacy information and improve the accuracy and effect of the recommendation system.Compared with other traditional recommendation algorithms,the algorithm proposed in this thesis can provide better recommendation effect and higher user satisfaction.However,to achieve true privacy protection,the algorithm needs to be improved further.How to improve the efficiency and scalability of the recommendation system on the premise of protecting user privacy,and how to improve the accuracy and robustness of the algorithm without increasing the computational complexity too much need to be further studied and explored.Firstly,adding differential privacy Gaussian noise will affect the accuracy and recommendation effect of the recommendation system.Although the privacy protection and recommendation accuracy can be balanced by adjusting noise parameters,excessive noise will lead to inaccurate recommendation results and reduce users’ trust in the recommendation system.Secondly,the use of SGD gradient perturbation processing increases computation and storage costs,because noise needs to be added to the training process and a large number of noise parameters need to be stored.Therefore,in practical applications,it is necessary to balance the relationship between privacy protection,computing and storage costs to achieve the best effect and performance.Gau+Ui-Label-SvD++algorithm proposed in this thesis is an effective privacy protection method,which can protect user privacy while improving the accuracy and recommendation effect of the recommendation system.The algorithm can adapt to different recommendation scenarios and data sets,and has strong flexibility and adaptability.