Knowledge Transfer In Cross Domain Recommender Systems Using Deep Latent Embeddings

Traditional recommender systems often face challenges such as the cold-start problem and data sparsity, especially in cross-domain scenarios. These issues limit their ability to provide accurate recommendations for new users or items with minimal interaction data, reducing their effectiveness of personalized applications. To address these challenges, a novel approach is proposed leveraging knowledge transfer in cross-domain recommender systems using Deep Latent Embeddings. The proposed method uses an autoencoder-decoder architecture to learn latent representations of users and items in a source domain and transfers to a target domain with sparse data. By mapping data from both domains into a shared latent space, the proposed approach facilitates effective knowledge sharing, enabling personalized recommendations even in data-limited domains. The proposed system is evaluated on three real-world datasets from Amazon benchmark datasets i.e. Books, Digital Music, and Movies & TV, under three cross-domain scenarios: (1) Books to Digital Music, (2) Movies to Digital Music, and (3) Books to Movies & TV. Empirical results demonstrate significant improvements in recommendation quality, mitigating cold-start and data sparsity issues. This research advances cross-domain recommender systems, providing practical solutions to enhance user satisfaction and engagement in recommendation applications.