Educated Rules for the Prediction of Human Mobility Patterns based on Sparse Social Media and Mobile Phone Data

摘要

Traditionally, mobility prediction at any level -as for example city, district regional or nationallevel- relies on household or individual level surveys. Nevertheless, the static information provisionfrom household/individual travel surveys for mobility prediction fails to capture the effects ofthe fast-evolving mobility trends, particularly today when individuals tend to relocate and changetheir mobility behavior more frequently than before. This paper presents techniques that handle automaticallyreal-time data from social sensing mechanisms (i.e., social media and mobile phones)to take advantage of the wide deployment of pervasive computing devices and the information exchangethrough them. The techniques are used to circumvent the shortcomings of traditional datasources and derive insights about the activity patterns of individuals for estimating their mobilitybehavior.In more detail, the automatic techniques comprise different rule-sets that process information aboutthe type of interactions (i.e., chatting via social media), the timing of interactions and the durationand re-currency of interactions to develop the mobility profiles of individuals and forecast theirmobility patterns during the week or the weekend. The techniques are validated against a fourmonthsample using information published on social networks by a set of users from the same city.The output of the techniques can be used for updating regularly the data provided by travel surveysor for developing tailored mobility forecasting models.