Negative Sampling in Collaborative Metric Learning: Cage-based Approach

Abstract

In this thesis, we start by pointing out the problem of a negative sampling (NS) strategy, denoted as nearest-NS, used in metric learning (ML)-based recommendation methods. The nearest-NS samples the items nearer to a user with higher probability among her unrated items. This could move her preferred items far away from her, thereby making the preferred items excluded from top-N recommendation. To address the problem, we first define a concept of a cage for a user, a region that contains the items highly likely preferred by her. Based on the concept, we propose a novel NS strategy, denoted as cage-based NS, that makes her preferred items rarely sampled as negative items. Through extensive experiments using four real-life datasets, we demonstrate that our cage-based NS strategy addresses successfully the problem of the nearest-NS strategy and outperforms consistently and significantly other NS strategies including the nearest-NS in terms of accuracy of recommendation.