Research on an explicit-implicit hybrid recommendation algorithm based on deep matrix factorization

Recommender systems benefit from combining explicit and implicit feedback to enrich user-item representations and mitigate data sparsity. Yet effectively utilizing the complementary nature of these feedback types to improve accuracy and generalization remains challenging. This study introduces a hybrid recommendation framework based on deep matrix factorization that integrates both feedback forms through a cross-attention mechanism. By modeling dynamic interactions between explicit ratings and implicit behavioral data, the approach facilitates a deeper understanding and fusion of their underlying relationships. Experimental evaluations reveal that this method outperforms traditional weighted or concatenated fusion strategies in capturing latent dependencies, resulting in superior recommendation quality.