基于多模态知识图谱的服装推荐算法OA

The existing clothing recommendation algorithms have problems with weak interpretability of models and incom-plete aggregation of user information.This article presents a novel clothing recommendation model called the multi-modal dual poli-cy knowledge graph(MDPKT),which enhances interpretability and aggregates user information more effectively.The MDPKT mod-el integrates multi-modal data—textual information like brand and clothing attributes,and visual data from clothing images—into a unified knowledge graph.It employs an improved KGAT model and a ResNet50 pre-training model to process textual and visual da-ta,respectively.A key feature the paper introduced in is the dual aggregation optimization strategy,which optimally selects the num-ber of aggregation hops for both users and items,improving information flow and model accuracy.The model is experimented on the IQON10 dataset and the IQON3000 dataset.The results show that this model can achieve better performance compared with other benchmark models.The improvement effect of the model proposed in this paper on the IQON10 dataset is more obvious.Compared with the SOTA model,MDPKT has significant improvements of 5.99%,5.12%,and 5.01%in AUC,Recall@20,and NDCG@20 re-spectively.On the IQON3000 dataset,these indicators have increased by 4.71%,8.1%,and 4.21%respectively.