基于区块链和联邦学习的隐私保护去中心化推荐系统OA

With the overload of Internet information,intelligent recommendation systems have emerged and are widely used to recommend products,content,and services to specific users.Yet,these systems require large amounts of user data to train their models.How to ensure the privacy and security of user data while maintaining model performance has become a pressing issue.Traditional recommendation systems suffer from problems such as data sparsity,and sharing raw data directly also can violate user privacy.This paper proposed a privacy-preserving decentralized federated learning recommendation system.It uti-lized the peer-to-peer network and immutable data storage features of blockchain to ensure data security and system decentrali-zation.In this system,user data was decomposed into private parameters(containing privacy information)and public parame-ters(containing item feature information)through matrix factorization.Users trained locally,kept their private parameters,and shared only the public parameters,which protected user privacy.Blockchain was introduced to coordinate the training process,where its leader aggregated local public parameters into a global public parameter.Users then downloaded and syn-chronized these parameters to conduct the next training round.Furthermore,it proposed a high-performance,low-consumption consensus mechanism based on a dynamic random seed algorithm,and the model's privacy-preserving performance was ana-lyzed.Experiments show that this system is superior to traditional centralized learning frameworks in terms of both privacy pro-tection and recommendation accuracy,while also offering strong scalability and practical usability.