Federated Learning With Adaptive Privacy Protection for Decentralised Intelligent Application

Deep learning enables recommender systems to capture intricate patterns and complex relationships between users and items across multiple domains. However, data sparsity remains a persistent challenge that degrades prediction quality. To address this, we propose a deep learning framework that integrates Deep Domain Adaptation to handle large-scale data and generate robust representations for cross-domain recommendation. Specifically, we introduce a novel deep neural architecture called Variational Autoencoder for Cross-Domain Recommendation (VAE‑CDR). The model employs shared encoders trained with a domain discriminator, ensuring that latent feature spaces for both users and items are maximally aligned between source and target domains. By enabling collaboration between domains, the system leverages mutual knowledge transfer, resulting in improved recommendation quality for both. Extensive experiments conducted on two real-world datasets demonstrate that the proposed method significantly outperforms state-of-the-art approaches, achieving higher accuracy and robustness in cross-domain recommendation tasks.