Exploring User Semantic Annotation from Trajectories in the Scenario of Shared Locations

Abstract. Over the past decade(s), collecting spatiotemporal data has become easier due to technological advancements and more user-friendly collection processes. Additionally, government agencies, companies, and open data projects have made general environmental data, such as points of interest or land use coverage, more freely available. Scientific studies can combine this spatiotemporal and non-spatial data to analyze different types of human mobility data. The results of these studies are relevant to transportation and urban planning, as similar information is typically collected by means of surveys. However, deriving relevant information from Global Navigation Satellite System (GNSS) trajectories remains challenging due to inaccuracies in the positioning and the unavailability of groundtruth information regarding individual user location semantics (e.g. home place, work place, leisure place or others). This work presents a semantic location annotation approach based on a Hidden Markov Model and the Viterbi optimization algorithm. The model includes location emissions to account for the general usage of a particular location. The annotations are applied to the clustered stop points that identify regions of special interest to individual users in a trajectory data set. The proposed approach demonstrates that the adapted Viterbi optimization can assign the most probable and meaningful semantic labels to the user’s sequences and provides insights on the underlying regions of special interest.