Hyperbolic Recommender Systems

  • 2018-09-05 19:30:54
  • Tran Dang Quang Vinh, Yi Tay, Shuai Zhang, Gao Cong, Xiao-Li Li
  • 23

Abstract

Many well-established recommender systems are based on representationlearning in Euclidean space. In these models, matching functions such as theEuclidean distance or inner product are typically used for computing similarityscores between user and item embeddings. This paper investigates the notion oflearning user and item representations in Hyperbolic space. In this paper, weargue that Hyperbolic space is more suitable for learning user-item embeddingsin the recommendation domain. Unlike Euclidean spaces, Hyperbolic spaces areintrinsically equipped to handle hierarchical structure, encouraged by itsproperty of exponentially increasing distances away from origin. We proposeHyperBPR (Hyperbolic Bayesian Personalized Ranking), a conceptually simple buthighly effective model for the task at hand. Our proposed HyperBPR not onlyoutperforms their Euclidean counterparts, but also achieves state-of-the-artperformance on multiple benchmark datasets, demonstrating the effectiveness ofpersonalized recommendation in Hyperbolic space.

 

Quick Read (beta)

loading the full paper ...