
Research Article
A Multi-behavior Recommendation Algorithm Based on Personalized Federated Learning
@INPROCEEDINGS{10.1007/978-3-031-54531-3_8, author={Zhongqin Bi and Yutang Duan and Weina Zhang and Meijing Shan}, title={A Multi-behavior Recommendation Algorithm Based on Personalized Federated Learning}, proceedings={Collaborative Computing: Networking, Applications and Worksharing. 19th EAI International Conference, CollaborateCom 2023, Corfu Island, Greece, October 4-6, 2023, Proceedings, Part III}, proceedings_a={COLLABORATECOM PART 3}, year={2024}, month={2}, keywords={Multi-behavior recommendation Privacy security Federated learning Personalized model Parameter encryption}, doi={10.1007/978-3-031-54531-3_8} }
- Zhongqin Bi
Yutang Duan
Weina Zhang
Meijing Shan
Year: 2024
A Multi-behavior Recommendation Algorithm Based on Personalized Federated Learning
COLLABORATECOM PART 3
Springer
DOI: 10.1007/978-3-031-54531-3_8
Abstract
Multi-behavior recommendation algorithms comprehensively use various types of interaction behaviors between users and items, such as clicking, collecting, purchasing, and commenting, to model user preferences and item features. It captures high-level interactions between users and items, and effectively alleviates the data sparsity problem in recommendation algorithms. However, most existing multi-behavior recommendation algorithms are mainly centralized learning models. User behavior data is collected and uploaded to the server to train recommendation model parameters, which poses a risk of data leakage and compromises user privacy. To address this problem, a multi-behavior recommendation algorithm based on the federated learning paradigm (FedMB) is proposed. This approach uses the federated learning framework to establish a separate model for each end device and utilizes the data of the end device for user-end model training, which improves the privacy and security of user data. To enhance privacy and security during parameters uploaded, all uploaded parameters will be encrypted, At the same time, the precedence chart is used to optimize the model parameters distributed by the server, thereby improving the recommendation quality of the overall model. Compared with that of the latest methods, our federated model achieves good performance on the three datasets.