Towards Efficient Hyperbolic Representation Learning for Recommender Systems

Abstract

Real-world relational data in recommender systems (RSs), particularly in domains such as e-commerce, often exhibits hierarchical structures, such as linking User → Product Category → Sub-Category → Item. Message-passing models such as Graph Neural Networks (GNNs), which propagate information across nodes, have been applied in most existing RS models. However, repeated message-passing steps in large-scale RS scenarios often leads to oversmoothing and oversquashing, where node embeddings collapse to similar vectors, reducing their discriminative power and degrading model performance. To address these issues, we propose HyperbRec, a hyperbolic neural network-based recommendation model that aims to preserve the hierarchical data structure and mitigate feature collapse. Experimental results on three real-world datasets show that HyperbRec achieves superior ranking performance and the highest memory and runtime efficiency compared to state-of-the-art baseline models.