Mining Trajectory Patterns with Point-of-Interest and Behavior-of-Interest

Abstract

Epidemiological investigation is one of the main means of controlling the outbreak of COVID-19. It has been proven to be effective, however, has a bottleneck that the infected person has to be questioned about his recent trajectory before any quarantine action could be taken, while sometimes trajectory information might not be timely and accurately obtained. In this paper, we propose an epidemiological investigation method which resort to artificial intelligence for extracting people’s preferences and social relationship from their historical trajectory patterns. Trajectory data used in our epidemiological investigation method may include time, location, Point-of-Interest (POI), as well as Behavior-of-Interest (BOI). All of these attributes in human’s trajectory are embedded into different channels in the proposed model and then fed into a classifier or a clusterer for serving different purposes. In our experiments, we applied the proposed method on a synthetic data set to conduct a classification task, and on a real data set for a clustering task. Both tasks confirm that the proposed method is effective and thus could be used to guide the preventive measures.