Federated Recommender Systems (FedRecs) enhance privacy protection by keeping user data on local devices, thereby reducing data interactions between users and the server. However, the server can still potentially infer the user’s relevant item set by analyzing the uploaded item gradients, which poses a threat to user privacy. In this paper, we propose a privacy-enhanced federated recommendation with explicit feedback named PriExRec to prevent user behavior data leakage in FedRecs. To reduce privacy risks while maintaining the utility of FedRecs, we first introduce an item mean imputation mechanism by combining Local Differential Privacy (LDP) with user rating habits. Empirical results reveal that the server can still successfully infer the user’s relevant item set by accumulating analysis of user-uploaded gradients over multiple rounds. To address this, we introduce a virtual client and use a consistent hash function to perturb the order of uploaded item gradients, making it difficult for the server to infer the user’s item set. Additionally, we adopt a federated feature compression mechanism, where both clients and servers use a compression matrix to handle high-dimensional features. Extensive experiments demonstrate the effectiveness of our proposed PriExRec in defending against interactive membership inference attacks (IMIA) while maintaining high recommendation quality.