Exploration of Collective Pattern to Improve Location Prediction of Mobile Phone Users

摘要

Location prediction based on cellularnetwork traces is a very challenging task due to therandomness of the human mobility patterns. Withthe help of the abundant social interaction datacontained in the cellular network, this paper focuson this question: How can knowing the location andthe assembled and dismissed behavior of my friendsbe used to more accurately predict my location?In this paper, we focus on how the collectiveeffect users' mobility. We notice an interesting rulethat users tend to stay around the places where theirfriends are denser. Those places where friends aremore crowned is chosen for a reason, either they arehaving a meeting there or they gatheredspontaneously because of sharing office building orother resources. With known locations of friends, we established a location prediction model to makefull use of those social information. In this model, we also introduced a measure of the ability of eachlocation that weather a gathering of friends here canattract the user. The result shows that this model did improvelocation prediction at the average of 4.91%, andmax improvement up to about 42.9%. And wesummarize the feature and the difference betweenthe kinds of users whose behavior is following hisfriends a lot from those user who move around more independently.