Spherical Graph Embedding for Item Retrieval in Recommendation System

Wenqiao Zhu, Yesheng Xu, Xin Huang, Qiyang Min, Xun Zhou

Abstract: One of the challenging problems in large-scale recommendation systems is to retrieve relevant candidates accurately and efficiently. Graph-based retrievals have been widely deployed in industrial recommendation systems. Previous graph-based methods depend on integrated graph infrastructures because of inherent data dependency in graph learning. However, it could be expensive to develop a graph infrastructure. In this paper, we present a simple and effective graph-based retrieval method, which does not need any graph infrastructures. We conduct extensive offline evaluations and online tests in a real-world recommendation system. The results show that the proposed method outperforms the existing methods. The source code of our algorithm is available online.