Cross-domain Recommendation using Gromov--Wasserstein distance

Abstract

A Cross-domain recommendation refers to a variety of item recommendation tasks that endeavor to suggest items to users across domains. This recommendation technique comprises various settings. In particular, this paper addresses a situation in which neither users nor items are shared in both domains. In such a scenario, it becomes challenging to apply traditional recommendation techniques since obtaining the similarity between users and items across domains is not straightforward. To tackle this problem, we propose a cross-domain recommendation approach based on the assumption that a group of users with shared preferences in one domain will also exhibit similar preferences in another domain. Our method utilizes the Gromov--Wasserstein distance to determine the similarity of users across domains. Through experiments conducted on multiple real-world data sets, we demonstrate the efficacy of our proposed method.