SanMove: next location recommendation via self-attention network

Purpose This paper aims to address the following issues: (1) most existing methods are based on recurrent network, which is time-consuming train long sequences due not allowing for full parallelism; (2) personalized preference generally considered reasonably; (3) rarely systematically studied how efficiently utilize various auxiliary information (e.g. user ID and time stamp) in trajectory data spatiotemporal relations among nonconsecutive locations. Design/methodology/approach The authors propose a novel self-attention network–based model named SanMove predict next location via capturing long- short-term mobility patterns of users. Specifically, uses module capture each user's long-term preference, can represent her preference. Meanwhile, use spatial-temporal guided noninvasive (STNOVA) exploit learn Findings evaluate two real-world datasets. experimental results demonstrate that only faster than state-of-the-art neural network (RNN) but also outperforms baselines prediction. Originality/value self-attention-based sequential trajectory, comprised learning modules. allows parallel processing trajectories improve efficiency. They an STNOVA transitions current trajectories. Moreover, used process historical order user. conduct extensive experiments check-in has fast training speed excellent performance compared with RNN-based