<p>Graph neural network, as a research hotspot in the field of recommendation system, can effectively deal with complex data structures. Although the existing recommendation model has made some progress, it still has obvious limitations. On the one hand, limited information about users and items may lead to insufficient accuracy. On the other hand, maintaining high accuracy while pursuing the diversity of recommended content is also a challenge. To address the above problems, we propose GDDRec, a diversified recommendation model based on graph diffusion, which introduces a self-gating mechanism that can accurately capture and implement the reinforcement of subtle information between users and items, thereby significantly enhancing the accuracy of recommended content. In addition, we further enrich the feature between users and items by generating diffusion heterogeneous graphs and integrating them with the graph aggregation module, while retaining the personalized interaction characteristics of users and improving the diversity of recommendations. To verify the validity of the GDDRec model, we conduct extensive experiments on two real-world datasets. The experimental results demonstrate that our model can significantly enhance recommendation diversity while ensuring accuracy, thus effectively balancing these two aspects to provide a richer and more personalized user experience.</p>