Enhancing recommendation systems with hybrid manifold regularized knowledge graph

Giang Ngo, Nhi N. Y. Vo

Abstract: Recent advances in graph neural networks have motivated the use of structured information in the form of a knowledge graph in recommendation systems. The most important challenges of real-world recommendation problems are the sparsity issues in which labeled user-item interaction data are sparse compared to the size of the knowledge graph. In this work, we propose a graph-based semi-supervised learning method namely Hybrid Manifold Regularized Knowledge Graph (HMR-KG), to tackle this problem. HMR-KG is an attentive knowledge graph neural network with a lightweight version of manifold regularization to enforce smoothness on the mapping function. Translation-based pre-trained embeddings are also used as initialization for the attentive knowledge graph neural networks. The method is evaluated using three public datasets. Experimental results show that our method not only outperforms state-of-the-art baselines but also yields more consistent results in scenarios with sparsely labeled data. In addition, our lightweight implementation successfully approximates the manifold regularization loss with substantially smaller time complexity.