A Graph Neural Recommendation With Role Information Enhancement

In recent years, User-Generated Content (UGC) has gradually become an important component of the internet content ecosystem. In order to improve the recommendation effect in UGC scenarios, existing work has successfully incorporated user role information into traditional recommendation models to assist in modeling different user behavior features. However, there are still three shortcomings. First, it ignores the differences in information richness when users assume different roles. Second, it fails to explicitly capture the higher-order connectivity information among nodes in the consumer-item and consumer-producer interaction graph. Third, the model parameters are optimized based on the negative sampling learning strategy. To address these issues, we propose a Dual-role Enhanced Graph Neural Recommendation (DERec) method to achieve personalized recommendation in UGC scenarios. Experimental results verify that the DERec outperforms six state-of-the-art baselines.