PRDG: Personalized Recommendation with Diversity Based on Graph Neural Networks

In recent years, recommendation systems have evolved to consider not only accuracy but also diversity and novelty when recommending item lists to users. Existing diversity models primarily focus on recommending the most diverse item lists to all users. However, not every user requires the maximum level of diversity in their recommendations. To address this, this paper proposes a PRDG (Personalized Recommendation with Diversified Graph Neural Networks based Recommender System) for recommendation. Unlike traditional single recommendation models, PRDG includes two advisors—one relevance-oriented and the other diversity-oriented. First, a user’s diversity scoring method is introduced to classify users. Then, different recommendation strategies are applied to different user to achieve personalized recommendations. Lastly, personalized negative sampling modules and re-ranking modules are proposed to enhance recommendation diversity and alleviate popularity bias. To validate the accuracy and diversity of the model, experiments are conducted on the Amazon-Beauty, Amazon-Music, and TaoBao datasets. The experimental results demonstrate that the proposed model significantly improves recommendation accuracy while maintaining comparable diversity to state-of-theart baselines.