Using Bert Embedding to improve memory-based collaborative filtering recommender systems

Abstract

The performance of memory-based collaborative filtering recommender systems will be severely affected when the users' item preference data is sparse. In this paper, we focus on solving this issue. Our idea is to use Bert Embedding to learn a new feature set, which is denser and more semantic, for re-representing users and items. In these new features, memory-based collaborative filtering recommender systems work more efficiently. The experiments are conducted on the Movielens 100K dataset and the Yahoo Webscope R4 dataset.