Abstract

In recent years there has been an explosion in the availability of data sets about colocation between individuals and connectivity with specific network infrastructure access points, from which user location can be inferred. These traces are usually collected through mobile devices equipped with short-range radio inter- faces, such as Bluetooth. Their potential is enormous as user movement data can be mapped onto the geographical space and the social interactions of individuals can be extrapolated from the colocation data. Quite interestingly, some of these data sets also contain a description of user profiles, such as the interests of the person, his/her age and gender and so on. In this paper we show that mobility and colocation information (i.e., social interactions) can be used to infer user interests by applying standard machine learning techniques. We evaluate a supervised and a semi-supervised technique using two different data sets containing information of interactions amongst people at conferences. We assume different degrees of available information for the inference problem and show that we are able to predict people?s interests with good accuracy also when only a small amount of information about user interests is available. While correlation of user interests with movement and proximity has already been investigated in social network research, this is the first work that uses machine learning to show this quantitatively.