Multi-source serialization cross-domain recommendation algorithm based on deep learning

Cross-domain recommendation is an effective approach to solve the cold start and data sparsity problems in recommendation systems. Sequential recommendation can model user behavior sequences and improve the accuracy of recommendation. Currently, few recommendation algorithms consider both aspects together, and most of them do not utilize multi-source information sufficiently. In view of this, this paper proposes a multi-source serialization cross-domain recommendation model, which fully considers the temporal and contextual relationships in two domains, and fuses multi-source information on the basis of achieving cross-domain recommendation tasks, and reinforce the embedding representation by fitting the interest forgetting function. Finally, use a Multilayer Perceptron as the mapping function to learn the nonlinear mapping relationship between the source domain and the target domain, subsequently enabling recommendations for new users in the target domain. On Amazon dataset, this model can significantly enhance the accuracy of recommendation.