Identifying points of interest from crowd-sourced trajectory data

Abstract

The high availability of GPS-enabled devices and easy use of smart phones have made it easy to generate and store location histories in trajectory data format. Large amounts of spatio-temporal data pertaining to an individual’s trajectories has given a rise to a variety of geographic information systems, and also brings us opportunities and challenges to determine valuable knowledge from these trajectories. Recent research efforts on trajectory dataset focus on identifying the labels of points of interest (POIs), recommending personalized POIs, deriving popular locations from GPS traces, etc. On the other hand, popular map services such as Google Maps, Bing Maps, etc. facilitate users to navigate different POIs. However, large number of POIs are yet to be covered by these map services. Comprehensive and up-to-date coverage of POIs are challenging because the number of POIs are large and the information related to POIs are always changing. Though some recent efforts have been made to identify POIs from check-in data or automatically identify POIs from images, these methods are costly and not suitable for changing circumstances. Crowdsourcing is a low-cost and efficient way to extract useful information from data acquired by crowd participants or volunteers. We could efficinetly find approximate location of a given point from the spatio-textual data collected from crowd sourcing. In this paper, we propose an approach to find an approximate location for a POI. The approach combines information from different sources such as the user’s texted input, movement and direction to find an approximate location for a POI. A large set of experiments on real datasets shows that our approach can perform significantly in terms of identifying a location.