Privacy-Preserving Next-Basket Recommendation Model

Existing next-basket recommendation methods rely on the centralized storage of user behavioral data for model training. However, due to the privacy-sensitive nature of such data, this approach carries a potential risk of privacy leakage. To address this issue, this paper proposes a federated learning-based next-basket recommendation model, in which user behavioral data is stored locally on users' devices and does not need to be uploaded to a central server. Communication between users and the server occurs solely through model parameters. To prevent the server or external attackers from identifying specific users via the uploaded parameters, the proposed method anonymizes user IDs. In addition, as model parameters may implicitly contain private information, local differential privacy is applied before uploading and distributing the parameters, offering stronger privacy protection. Extensive experiments conducted on public datasets demonstrate that the proposed method can effectively preserve user privacy while maintaining strong model performance.