Content-based Art Recommendation Using Multimodal Graph Neural Networks

Graph Neural Network (GNN)-based recommendation systems have become very popular in recent years. Their popularity stems from the fact that nodes can access higher-order neighbor information and there are well-designed algorithms for embedding, message passing, and propagation. In this paper, we present the design of a GNN-based recommendation system on a novel data set collected from field research. Designed for an endangered performing art form, the recommendation system uses multimodal text and image data to suggest similar panels (paintings) to the end-user of the system. We show that multimodal data is particularly helpful in graph representation learning, message passing, and propagation and the panels recommended to users using content-based features are similar to those obtained from systems using user-item bipartite graphs such as MMGCN and GRCN.