Discrepancy Learning Guided Hierarchical Fusion Network for Multi-modal Recommendation

Yuzhuo Dang, Zhiqiang Pan, Xin Zhang, Wanyu Chen, Fei Cai, Honghui Chen

Abstract: Highlights•A novel multi-modal fusion framework for recommender systems is proposed.•Complementarity between early fusion and late fusion is identified and utilized.•A popularity-aware pruning strategy is proposed to remove the noisy interactions.•Extensive experiments on three datasets demonstrate the superiority of DGHNet.