Research On Location Prediction And Recommendation Method Based On GPS Trajectory Data

LBS(Location Based Service)is a kind of information service,which provided to users about the geographical positions located by mobile divices and wireless network.With the popularity of smart devices,it is much easier to access to the information provided by the geographical locations.Besides,there is a wealth of information in the location data,such as users' interests,users' hobbies and user's behavior pattern.Among all the techniques about LBS,the location prediction methods and location recommendation methods are two important technologies.An effective location prediction or recommendation can make the users have good experience,and has high practical value and application value.The main works of this dissertation are shown as follows:(1)Improving the performance of sparse trajectory points-Location representation method based on road networkThis dissertation proposes a location representation method based on the road network,which uses a polygon to approximate a location area.The polygon is decomposed into a set of triangles,which simplifies the composition of polygon.The experimental results on GeoLife and T-Drive datasets showed that the proposed method could effectively represent an unknown region in the absence of a small number of trajectories.(2)A position prediction model based on continuous time information--Markov model based on continuous time seriesUsing traditional Markov model will lose the time information or users' transferring information according to the users' location sequence or time sequence.To solve this problem,this dissertation proposes a new Markov model based on continuous time series.The model find out the time point where the users maybe transfer between two different locations by establishing Gaussian Mixture Model based on users' transfer time point.The model not only preserved the time information,but also save the transferring information.The experimental results on Geo Life dataset showed that the proposed method can not only accurately predict the user's location at a certain time,but also improve the accuracy of location prediction.(3)A method to improve the performance of location recommendation – Location recommendation based on time decay factor.The traditional collaborative filtering method and singular value decomposition method have been proved to be effective in practice.However,the users' location movement behavior often change with time,so this dissertation proposes a new location recommendation based on time decaying factor.First,we divide the dataset by different time and give a decaying factor to the divided dataset,and then we establish a loss function and minimize the loss function to study the decaying factor.Finally,we use the decaying factor to weight the recommendation results produced by different time dataset.The experimental results on Gowalla dataset showed that the proposed method has higher precision value and recall value compared with the traditional method,and has better recommendation effect.