An enhanced recommender system based on heterogeneous graph link prediction

Graph-based data has gained popularity in various applications, including social networks, recommendation systems, and knowledge graphs. Detecting missing links between nodes is a critical challenge in processing graph-based data, as it directly impacts the performance of these applications. The framework for inductive representation learning on large graphs called GraphSAGE method is commonly used for identifying node embeddings, but it has limitations in terms of sensitivity to hyperparameters and difficulties in processing large graphs and capturing global graph information. To address these issues, we propose an improved GraphSAGE architecture for link prediction. Our approach leverages advanced aggregation functions and diverse neural network architectures. Through extensive experimentation on benchmark datasets, we demonstrate the superiority of our method in terms of link prediction accuracy compared to standard GraphSAGE. Additionally, by integrating clustering techniques, we develop a robust recommender system that outperforms state-of-the-art techniques. Our findings highlight the practical potential of our methodology for recommendation tasks and underline its significance in graph-based data processing.

Automated summary

Graph-based data is widely used in various applications, and detecting missing links between nodes is a critical challenge. We propose an improved GraphSAGE architecture for link prediction, leveraging advanced aggregation functions and diverse neural network architectures. Our method outperforms standard GraphSAGE in terms of link prediction accuracy, and integrating clustering techniques further enhances its performance in recommender systems.