Dual contrastive enhancement-based multimodal recommendation

Multi-modal recommendation systems are widely applied in user service fields such as e-commerce and social platforms. They have become a key method to improve user experience and promote personalized recommendations. However, existing recommendation systems face challenges such as long tail distribution, data sparsity, and semantic mismatch between modalities. These challenges limit the accuracy and robustness of recommendations and reduce user satisfaction. To address these issues, a multi-modal recommendation system DCE-MMR is proposed based on dual contrastive enhancement in this paper. Firstly, we utilize intra-modal contrastive learning to improve the expression of item features. The impact of long tail distribution and data sparsity is reduced by item feature optimization with contrastive learning; Secondly, inter-modal contrastive learning is adopted to align the features of different modalities and address the problem of semantic misalignment between modalities. Finally, the features from different modalities are fused with Cross Attention mechanism to enhance the feature representation. Experiments were conducted on two recommendation datasets. The experimental results show that the proposed method has significant improvements compared to existing methods, especially in terms of recommendation accuracy for processing long tail data and sparse data.