文献类型：期刊文献

英文题名：Federated Recommendation Algorithm Based on Model Comparison

作者：Huang, Rui[1]; Guo, Weibin[1]

机构：[1] College of Information Science and Engineering, East China University of Science and Technology, Shanghai, China

年份：2023

外文期刊名：2023 International Conference on Data Science and Network Security, ICDSNS 2023

收录：EI(收录号：20234214881388)

语种：英文

外文关键词：Behavioral research - Learning algorithms - Learning systems - Recommender systems

摘要：The recommendation system uses user historical behavior data and user personal information to recommend goods, movies, music, etc. However, user data is easy to leak during transmission, which threatens user privacy. Federated learning is a technique that enables multiple parties to jointly train machine learning models without exchanging local data. However, the actual distribution of user data is very uneven, which is a key challenge for federated learning. In order to solve the problem of user privacy protection and non-independent distribution of data, a federated learning recommendation algorithm based on model comparison (MC-FedRec) is proposed. The algorithm optimizes the local model learning effect by improving the local training part of FedAvg and using the similarity between each model representation. In the experiment, three public recommended datasets were used. The datasets were segmented by the Dirichlet distribution and assigned to each participant to simulate the non-independent distribution of the data. Experimental results show that the proposed algorithm is superior to other baseline federated learning algorithms in recommendation accuracy, communication efficiency and robustness. ? 2023 IEEE.