A Hierarchical Multi-modal Content-Based approach to Graph-based Recommendation System

This study offers a novel architecture based on graphs and is intended to improve the representation of user-item interactions in recommendation systems. Due to the utilisation of bipartite graph; decomposition and heterogeneous message propagation; this technique successfully captures intricate user-item interactions. Systems uses graph convolutional networks (GCN) and meta-path-based graph neural networks (GNN) to store features that make the user-item interaction space more dependable. A GCN decoder reconstructs the user-item interaction network to enhance system understanding, while a multi-class classifier enables personalised recommendations based on the user’s preferences. A combined loss propagation mechanism, advanced similarity matching, and link prediction methods improve the recommendation process. This makes the digital platform more accurate and customer-satisfying. The proposed method, after validation on real-world video datasets surpassed the recent techniques.