Cross-fusion activates deep modal integration for multimedia recommendation

Recommendation systems play a significant role in information presentation and research. In particular, goods recommendations for consumers should match consumer psychology, speed up product search, and improve the efficiency of product transactions. Online platforms provide product information and interactive information between customers and products. However, the interactive modeling effect of the existing multimedia algorithms on this information must be improved, for instance, by deeply integrating product and interactive information. Accordingly, we propose a cross-fusion-activated multi-modal (CFMM) integration method for recommender systems to achieve deep fusion of product and user information. This method adds a cross-fusion module to fuse the features of different modalities through deep-feature fusion. A fusion loss function is further proposed to improve the recommendation performance of the network. Extensive experiments were conducted on three real-world datasets along with multiple ablation studies to illustrate the effects of the different modules. The experimental results show that the proposed method exhibits better recommendation performance, providing a maximum improvement of 3.8% in the recommendation performance metrics Recall@20, NDCG@20, and Precision@20 in comparisons with existing algorithms. This method realizes a deeper integration of multimodal information; however, the performance can be further improved by extending the multimodal information interaction algorithm to include product and user information.