Title: Moving object location prediction based on a graph neural network with temporal attention

Authors: Jun Qian; Yubao Wu

Addresses: School of Information Technology, Nanjing Forest Police College, Nanjing, 210023, China ' School of Information Technology, Nanjing Forest Police College, Nanjing, 210023, China

Abstract: Predicting the location of moving objects is a crucial component of location-based services that provide decision-making support for applications such as smart transportation, etc. Long-term dependencies on daily and weekly cycles are a part of individual mobility patterns, where user behaviour is heterogeneous within a cycle but highly homogeneous across different cycles. Based on this observation, a moving object location prediction model is proposed based on a graph neural network with temporal attention (GNN-TA). First, this model proposes a location-distributed representation method based on a graph neural network (GNN-LDRM). This method is used to obtain low-dimensional location embedding vectors that contain potential correlations by reducing high-dimensional vectors. Then, bidirectional long short-term memory networks and multi-head self-attention mechanisms capture time and space information. Finally, a personalised temporal attention mechanism is constructed to capture users' long/short-term mobility patterns to predict moving object location. Experiments on real datasets show that the GNN-TA model has a significantly improved prediction accuracy compared with traditional methods.

Keywords: location prediction; graph neural network; GNN; temporal attention; moving object.

DOI: 10.1504/IJSN.2023.134133

International Journal of Security and Networks, 2023 Vol.18 No.3, pp.153 - 164

Received: 26 May 2023
Accepted: 28 May 2023
Published online: 11 Oct 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article