Research Of Location Prediction Technology Using Semantic Information Of Locations

In recent years,using user's history trajectory to mine the user's mobile track position information and predict user's next location has become a hot topic.While most prediction techniques only use the geographic and temporal features without semantic features,we propose the research on location prediction technology using semantic information of locations.The main work of this paper is as follows:First of all,we implement a semantic location mining method,which is mainly responsible for the conversion of the original trajectory of the user's GPS to the geographical and semantic trajectory.It includes two clustering processes.The first process converts the GPS trajectory to the stop point sequence by setting the distance and time threshold.The second process uses the density-based clustering function to cluster the stop points into a significant geographical position,and transfer geographical location to semantic location by using method of reversing geocoding.Semantic location mining is the first step in location prediction.Secondly,we implement a method based on moving frequent pattern,which can improve the accuracy of location prediction by using frequent sequential pattern mining algorithms.This method uses multi-information of locations including geographic,temporal and semantic information.The experimental results show that,compared with the location prediction algorithm based on the geographic and temporal attributes,the accuracy of the semantic location prediction based on moving frequent pattern improves by 17.89%.Finally,we implement an adaptive multi-order Markov model for location prediction using geographic,temporal and semantic information of locations.This method can adaptively select the order of Markov model by the user's current trajectory sequence and the historical trajectory pattern tree.It overcomes two shortcomings of the uncertainty caused by the low order and low forecast coverage caused by the high order.The experimental results show that the accuracy of the adaptive multi-order Markov model improves by 8.96%when compared with the multi-order Markov model.