Se-Transformer: A Multimodal Recommendation System Based on Channel Attention Mechanism

Currently, multimodal recommendation has gradually become a key focus. How to effectively utilize the information between modalities to achieve multimodal recommendation is a crucial factor in retaining customers. In multimodal recommendation, how to effectively enhance the feature extraction ability of the recommendation system model has become a key issue. This study intends to adopt a concatenated fusion method of Se-net and Transformer to comprehensively improve the model’s feature extraction ability and achieve multimodal recommendation. Experiments show that the Senet model can effectively enhance the feature extraction ability of the Transformer model through the channel attention mechanism. A recommendation system was implemented on the Amazon baby dataset, achieving better recall and normalized discounted cumulative gain compared to baseline models such as mamba, GNN, KKNCNF, and MVGAE.