Diverse Graph Convolutional Process for Recommendation System

In recent years, while the diversity of recommendation systems is closely linked to user satisfaction, it has often been overlooked. Most studies focus solely on recommendation accuracy, considering diversity only in the candidate generation stage. To tackle this challenge, this paper introduces a method that integrates diversity into the candidate generation phase. by improving Graph Convolutional Networks (GCNs). Although GCNs have shown remarkable performance in enhancing recommendation accuracy, they lack sufficient consideration for diversity in existing research. To this end, we introduce rebalanced neighbor discovery, category-enhanced negative sampling, and a fast computation method based on determinantal point processes to optimize the trade-off between diversity and accuracy. Experiments on real-world datasets show that the proposed method effectively improves recommendation diversity while maintaining high accuracy.