关键词: 旅伴同伴, 自编码器, 时空信息, STR算法, 注意力机制

Abstract: With the widespread adoption of mobile devices, today’s location tracking systems are producing tremendous amounts of trajectory data on a continuous basis. The ability to discover moving objects that travel together (i.e., traveling companions) from their respective trajectories is desirable for many applications, including intelligent transportation systems and intelligent advertising. Existing algorithms are either based on pattern mining methods that define a particular pattern of traveling companions or based on representation learning methods that learn similar representations for similar trajectories. The former method suffers from the pairwise point-matching problem, and the latter often ignores the temporal proximity between trajectories. In this work, we propose a deep representation learning model using autoencoders, namely Mean-Attn (Mean-Attention) , for the discovery of traveling companions. Mean-Attn collectively injects spatial and temporal information into its input embeddings using skip-gram and positional encoding techniques, respectively. In addition, our model encourages trajectories to learn from their neighbors by leveraging the sort-tile-recursive (STR) algorithm as well as the mean operation and global attention mechanisms. After obtaining the representations from the encoder, we run DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to cluster the representations and find traveling companions. Experimental results suggest that Mean-Attn performs better than the state-of-the-art data mining and deep learning algorithms for locating traveling companions.

Key words: traveling companion, autoencoders, spatiotemporal information, sort-tile-recursive algorithm, attention mechanism