Research Area:  Machine Learning

Abstract:

Existing neural-network based recommender systems usually first employ matrix embedding (ME)as a pre-process to learn users and items representations (latent vectors)to make accurate Top-k recommendations. However, most NN-RSs focus on accuracy by building representations from the direct user-item interactions, while ignoring the underlying relatedness between users and items, which is an ideological drawback. On the other hand, ME models directly employ inner products as a default loss function metric, which is amethodological drawback. In this paper, we propose a supervised collaborative representation learning model - Magnetic MetricLearning (MML) - to map users and items into a unified latent vector space which can enhance the representation learning.MML utilizes dual triplets to model not only the observed relationships between users and items, but also the underlying relationships between users, as well as items to overcome the ideological drawback. Specifically, a modified metric-based dual loss function is proposed to gather similar entities and disperse the dissimilar ones. We conduct extensive experiments on four real-world datasets with large item space. The results demonstrate that MML can learn a proper unified latent space for representations from the user-item matrix with high accuracy and effectiveness.

Keywords:  

Author(s) Name:  Yuanbo Xu; En Wang; Yongjian Yang; Yi Chang

Publisher name:  IEEE

DOI:  10.1109/TKDE.2021.3054782

Volume Information:   Page(s): 1 - 1

Paper Link:   https://ieeexplore.ieee.org/abstract/document/9336271