On the Predictability of a User's Next Check-in Using Data from Different Social Networks

Abstract

Predicting a person's whereabouts is important in several scenarios. However, it is hard to obtain data that reliably reflects users' mobility patterns. This difficulty has led researchers to use social media data as a proxy to understand and predict human mobility. It has been shown, however, that such data is inherently biased and error-prone, and that such drawbacks may produce sub-par mobility prediction models. In a more narrow context, researchers have used social media data to predict users' check-in patterns. A common approach to alleviate the biases in social media data is to use more than one data source. We here show, however, that the use of data from different social networks does not necessarily increase the predictability of a person next check-in. Our experiments indicate that this result is due to how and where people use different social networks, and that user behavioral characteristics play an important role on the predictability of the next check-in.