Cross-Domain Recommendation: Leveraging Semantic Alignment and User Clustering to Address Data Sparsity

Cross-domain recommender systems can address data sparsity by leveraging information from a data-rich domain to improve recommendations in a data-sparse domain. In this study, we consider two distinct domains that share common members but have different items. We propose a new approach to enhance recommendation accuracy in the sparse domain by utilizing semantic alignments and clustering techniques. We begin the process by aligning the domains using shared semantic information between them. After establishing this semantic alignment, we apply clustering techniques to group similar users within each domain. These user clusters are then aligned across domains, allowing us to transfer knowledge from the richer domain’s clusters to the sparser domain. By effectively bridging the gap between the domains, our method can enhance the accuracy of the recommendation. We have evaluated the performance of our proposed approach on the Amazon Movies and Amazon Books datasets.