With the development of the Internet, the amount of information on the network has grown exponentially, leading to the problem of information overload. The harm of information overload lies in the difficulty for users to obtain the desired information from a massive amount of data. As one of the typical big data applications, recommendation systems aim to extract data using recommendation algorithms and provide users with suggestions for potentially interesting items. However, the risk of personal data leakage arises as a result. To address this issue, this paper proposes a recommendation algorithm based on collaborative filtering matrix factorization and applies horizontal federated learning to model separation. By combining federated learning with recommendation systems, the application of federated recommendation algorithms is achieved. Comparative experiments and result analysis are conducted on the MovieLens dataset to compare the performance of federated recommendation algorithms with non-federated ones. The experimental results indicate that this approach ensures both recommendation accuracy and the security of user privacy data. Finally, we propose an improved solution based on the optimal replacement algorithm for communication between federated learning servers and clients, aiming to enhance the efficiency of servers in handling multiple user requests within a short period of time.