A Complementary Product Recommendation Method Based on Modality Graph

Complementary product recommendation is to recommend products for users that can be used together. Existing product recommendation methods consider the simple features of products' image and text without further capturing the depth features of image and text, and they mainly place emphasis on exploring the intra-modal complementary relationships between products, but ignore the importance of the inter-modal complementary relationships. In addition, most of these methods achieve cross-modal fusion by directly splicing multiple modalities, thereby probably causing sub-optimal performance. Based on this situation, this paper proposes a Complementary Product Recommendation Model Based on Modality Graph, namely, (CPRMG). The model uses multi-modal information of products as input, further enriches the representation of image and text by gated self-attentive module, then combines the intra-modal complementarity with the inter-modal complementarity, and finally fuses them based on Transformer encoder to learn the complex relationship between products and candidates for complementary product recommendation. Experimental results on the Amazon dataset demonstrate the CPRMG model proposed in this paper outperforms other baselines in all metrics. The experiments show that the CPRMG model can recommend complementary products for users more accurately.