Multiview comparison recommendation model integrating interactive intention

Recommendation models driven by graph neural networks (GNNs) face two major challenges: first, the lack of supervision impedes the accurate capture of the interaction details between users and items; second, the inadequate understanding of the complex relationships between users and items compromises the interpretability of recommendation results. To address these challenges, this paper proposes MCRII, a multi-view contrastive recommendation model that integrates user interaction intentions. The model incorporates structured information from the knowledge graph, combining global structural views with local perspectives on user intent, coordination, and collaboration, thus revealing the intrinsic relationships between users and items. Through multi-view contrastive learning, MCRII enhances the discriminative power of user-item representations and improves the explainability of recommendations. By integrating contrastive learning with multi-task learning, MCRII makes significant strides in self-supervision, contrastive learning, and intent learning tasks. Experimental results show that MCRII outperforms baseline models in CTR prediction on the Amazon-Book and Yelp2018 datasets.