MobTCast: Leveraging Auxiliary Trajectory Forecasting for Human Mobility Prediction

• DOI: 10.48550/arxiv.2110.01401
• Publication Date: 2021-01-01
• Authors: Hao Xue; Flora D. Salim; Yongli Ren; Nuria Oliver
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; Machine Learning (cs.LG)

Human mobility prediction is a core functionality in many location-based services and applications. However, due to the sparsity of data, it not an easy task predict future POIs (place-of-interests) that are going be visited. In this paper, we propose MobTCast, Transformer-based context-aware network for prediction. Specifically, explore influence four types context prediction: temporal, semantic, social geographical contexts. We first design base feature extractor using Transformer architecture, which takes both history POI sequence semantic information as input. It handles temporal Based on connections user, employ self-attention module model context. Furthermore, unlike existing methods, introduce location branch MobTCast auxiliary next location. Intuitively, distance between predicted from should close possible. To reflect relation, consistency loss further improve performance. our experimental results, outperforms other state-of-the-art methods. Our approach illustrates value including different