Purpose This paper aims to address the following issues: (1) most existing methods are based on recurrent network, which is time-consuming to train long sequences due to not allowing for full parallelism; (2) personalized preference generally are not considered reasonably; (3) existing methods...