We have been developing a simple friendship mobility model that captures the essence of how a group of friends travels together or move from one location to another. The idea is pretty simple, but the results are powerful.
First, we collected data from location-based social networks, specifically GoWalla. Second, we used a Marko model where checkins represented states and the transitional probabilities of going from one state to another was empirically defined by the training dataset. Since GoWalla provides not only the mobility traces of individual users through a process known as checking in, but GoWalla also provides the friendship topology and other cool stuff too! Therefore, each user gets his or her own empirical Markov Model, and the complete Markov Model consists of each one. Once we have the complete Markov Model, we use Miller’s coordinate system to convert latitude/longitude into a Cartesian system that preserve distances.
This is how a set of friends travels together. Implications? Huge. For instance, take the popular Random Waypoint Model that has dominated simulations in the networking research community and replace that with the Friendship Mobility Model. Or study the migration of a population during a natural disaster; e.g, the recent nuclear disaster in Japan. Enough Said.
The paper, “Using Location-Based Social Networks to Simulate Human Mobility for Mobile Networks” is currently under review for NetSciCom 2012, in conjunction with InfoCom. More results will be present and datasets/code will be publicly available and disseminated by an agreement with IRB @ RPI under the protocol #1125 entitled, “Data Infrastructure for Complex Social Networks.”
Of course, we aren’t the only one or the first to study human mobility. Check out the awesome work of these folks. http://www.barabasilab.com/ http://cs.stanford.edu/people/jure/
