Research And Implementation Of Recommendation Algorithm Based On Differential Privacy Protection

With the rapid development and popularization of internet technology,information overload has become a pervasive issue in today’s society.In response to this problem,recommendation systems,as an adaptive information filtering and personalized recommendation technology,have become an essential component of internet services and commercial applications.Based on machine learning and data mining techniques,recommendation systems analyze users’ historical behavior,interests,and preferences to automatically recommend relevant products,services,or information,thereby assisting users in finding the beneficial information they need amidst the clutter and enhancing information retrieval efficiency and satisfaction.While recommendation systems have grown in popularity,they also pose significant privacy risks for users.Training recommendation models requires extensive user behavior data,which may comprise private user data such as medical diagnoses or purchasing history.Although privacy issues in recommendation systems are increasingly prominent,research in this area remains limited thus far.Addressing the above problems,this paper presents two differential privacy protection techniques—the centralized differential privacy deep learning recommendation algorithm PrivANCF and the localized differential privacy deep recommendation algorithm LPGCF—for trusted and untrusted third-party service center scenarios,respectively.First,in the case of trusted third-party service centers,the PrivANCF model is employed.This model implements differential privacy via adaptive differential privacy random gradient descent and utilizes the selfattention mechanism to thoroughly mine user-item interaction feature information.Combining the best of both approaches,this model achieves an excellent balance between privacy and accuracy.Second,for scenarios involving untrusted third-party service centers,the LPGCF model is proposed.This model uses the multi-bit mechanism to derive high-dimensional feature data from clients’ mobile phones and employs the graph neural network-based high-order embedding propagation layer to effectively extract interactions between users and items.As a result,the model achieves both impressive privacy and recommendation accuracy.Lastly,by innovating based on the two algorithms,a privacy-adaptive movie recommendation system is devised,including system requirement analysis,system architecture design,implementation technology,and application effectiveness.By presenting the frontend implementation webpage for the system’s main functions,the practical applications of the system are demonstrated intuitively.This paper conducts experiments on multiple public datasets to verify that both the PrivANCF and LPGCF models,without sacrificing too much accuracy,satisfy differential privacy requirements,achieving a desirable balance between privacy and accuracy.Lastly,the privacy-adaptive movie recommendation system,which incorporates these two algorithms,effectively demonstrates the applied implementation of the algorithms.