Prediction of Destinations and Routes in Urban Trips with Automated Identification of Place Types and Stay Points

Predicting the destination and the route that someone is likely to take is useful for various purposes, such as to prevent people from going through congested routes. Most of existing approaches to this prediction problem only consider geographic patterns within their models, although this appears to be not enough for creating a robust predictor. This paper proposes an approach to improving the task of predicting route and destination which makes use of further semantic information associated with destinations and routes, apart from location patterns. Our model does not require user's active interaction and is able to automatically identify stay points (i.e., places users visit) and type of places. We evaluated our model with real world data collected from users’ smartphones and obtained promising results.