Data Driven Human Movement Trend Modeling

In recent years,with the wide spread of the GPS enabled electronic devices,and the check-in functionality provided by the location based social network services,we can collect large amount of precise location data of human movements.By modeling those data,it makes the analysis on human movement trends come to true.In our work,we model the human movement trends based on the data of movement history,and then we use the model to predict the location of the user.Prediction on user's location is a fundamental problem of importance in the field of location based service.In the base resource management,location based personalized recommendations,intelligent traffic and city planning,there are many promising applications based on location prediction.With the movement history data only containing the geometry information and temporal information,we propose to model the temporal and spatial mobility patterns of human movement with input-output Hidden Markov Model.In order to account for data missing,we introduce the dummy state into the construction of Markov data chain.With dummy states,we avoid generating those transitions between places not really happened in the case of directly cascading data.We also propose three predictors based on the temporal dynamic transition probability,temporal transition probability and spatial transition probability respectively.We finally predict locations with the combination of these three predictors.With the movement history data not only containing the geometry information and temporal information but also the semantic information,we propose the semantic temporal spatial Hidden Markov Model by extending the input-output Hidden Markov Model to model the mobility behavior of human movements.According to whether we have the knowledge on the semantic information of the current location,we propose two different predictors to make the prediction of locations respectively.The experimental results show that our methods obtain better prediction accuracy than several state-of-the-art works which proves the effectiveness of our methods.By considering both the temporal and spatial mobility patterns of human movements,we can model the human movement trends better and gain higher prediction accuracy.And taking semantic mobility pattern into account can further improve the performance of prediction accuracy.