Research On Location Prediction Based On Multi-characteristics Of Trajectories

In recent years,with mobile localization technology maturing and popular,location based service(LBS)receives extensive attention.Location prediction is an important part of LBS,which is widely applied.At present,the common methods of location prediction are based on Markov model or based on frequent patterns mining,in which many problems exist.For example,they take more time to mine frequent trajectories,prediction is difficult when a new mobile pattern appears,and the order number of Markov model is difficult to determine.Those methods are mostly based on the spatial characteristic or spatial-temporal characteristics of the trajectories,without taking the semantic characteristic reflecting users'behaviors into account.To solve this problem,this thesis puts forward a new location prediction method based on spatial,temporal and semantic characteristics of trajectories.In order to improve the accuracy of prediction,this thesis considers the user behavior and the range where the users move respectively,and predicts users' behaviors based on the clustering behavior.And then the predicted behavior is combined with the range to get the final prediction result.By preserving the multi-step transition probability matrix between areas,repeated calculation is avoided and the efficiency of prediction is improved.The main contributions of this thesis are summarized as follows.First,the thesis proposes a new method of location prediction based on the spatial,temporal and semantic characteristics of trajectories,which can not only predict the location of a user,but also obtain the user's behavior.Second,to solve the problems of extracting stay points existing in traditional methods,this thesis proposes a new method to obtain visited points considering distance,time,and direction at the same time.After obtaining meaningful visited points,visited places are extracted.Moreover,to solve the problems of partitioning map by grid existing in traditional location prediction,this thesis proposes a new map-partitioning scheme based on intersections and Voronoi diagram.Third,to solve the problem that the current location prediction methods do not take into account users' behaviors,this thesis proposes a semantic prediction model based on users,behaviors.The semantic sequence similarity is computed and clustered to predict users'behaviors.Besides,this thesis proposes the spatial prediction model to predict users'locations based on Bayes theorem and Markov model.In the spatial prediction model,transition probability matrix is obtained by considering the detour distance,which avoids the data sparseness problem.Finally,this thesis puts forward a semantic matching strategy to predict users' behaviors based on semantic pattern trees.The proposed strategy can achieve full and partial semantic matching at the same time,which improves the accuracy of location prediction.