Towards an Affective Semantic Trajectory Generator (ASTG)
Antonios Karatzoglou, Markus Szarvas, Michael Beigl · 2018
Trajectory modelling, trajectory analysis and trajectory prediction have become very important tools in the hands of mobile service providers, whether in respect to resource management (e.g., mobile network management), or to building intelligent, context-aware mobile applications. Most of the existing modelling approaches are highly data-driven. For this reason, the need of large, high-quality datasets has become enormous in the recent years. It is very costly and time-consuming to collect real-world data such as human trajectory data. Moreover, new privacy laws and restrictions make it even more harder. Thus, data turned into a bottleneck for algorithm developers of all kinds. Synthetic data generators provide a solution for this problem. There exists a variety of approaches for producing synthetic trajectories and many extra features have been investigated such as the transportation mode, the proximity to friends and the activity to name but a few. However, none of them has explored the use of psychological features, such as the personality and the emotional state of the users. In this work, we try to give insight into the impact of the aforementioned features on the generation process of (semantic) location trajectories. For this purpose, we designed a novel multi-agent synthetic trajectory generator that takes, among others, these features explicitly into account. We refer to it as Affective Semantic Trajectory Generator (ASTG). In order to evaluate our approach and the use of personality and emotions, we compared the produced trajectories with the outcome of two large-scale studies (> 25.000 participants each) conducted in Germany and Chicago, USA in 2008. It can be shown that dynamic data, such as emotions, can lead to a better performance, a fact that makes ASTG particularly interesting for further investigation.