Graph Convolutional Networks for Designing Collaborative Filtering-Based Health Recommender Systems

Recommender systems provide useful item suggestions (products or services) to users as part of their decision-making processes. The effectiveness of recommender systems is now clearly confirmed in various fields of application (e.g., YouTube, Amazon, Facebook, ResearchGate). In the literature, many research works have addressed the application of recommendations in the field of health in what are called health recommender systems (HRS). HRS is an innovative alternative when it comes to providing information to help doctors in the diagnosis/treatment of diseases, as well as helping patients with recommendations on how to maintain their well-being. However, the proposed development approaches in this field are limited to traditional models that lack the accuracy and effectiveness, which are vital in healthcare. This paper presents a design model for collaborative filtering-based health recommender systems using graph neural networks (GNN) via its promising Graph Convolutional Network (GCN) architecture. In this model, the convolution layer works with a simplified and efficient GCN algorithm named LightGCN. GCN-based methods are among the new cutting-edge approaches in recommender systems, and LightGCN has proven its superiority in recommendation accuracy.