Cross-Domain Recommendation Framework for Enhanced Personalization through Contrastive Learning

Abstract To tackle the persistent data sparsity issue in recommendation systems (RSs), cross-domain recommendation has emerged as an effective solution by leveraging data from richer domains to enhance performance in sparser ones. Our proposed framework not only addresses data limitations but also ensures that the recommendation process remains fair, transparent, and user-centered. By employing contrastive learning and multi-head attention, proposed framework creates uniform and unbiased user/item embeddings, counteracting issues of popularity bias and embedding inconsistencies. A critical aspect of this framework is its behaviour encoder, designed to learn user preferences through behavioural data such as clicks, reviews and sentiments ensuring a personalized experience for users even in domains with sparse data. To maintain fairness and transparency, the system adaptively balances user domain specific and domain shared features based on user preferences in target domain.