Research On User Location Prediction System Based On Cellular Data

The rapid development of modern positioning technology enables operators to obtain the geographic location information of large-scale users through smart mobile devices.At the same time,the mining of location information can realize a variety of location-based services and create benefits for operators.In location-based service scenario,location prediction,as the basic research problem,plays an important role in understanding the user space movement pattern.In this paper,the user location prediction system is built based on cellular data.In this location prediction system,in order to achieve more accurate decision-making and improve user experience,how to deal with the complex scenes of space-time trajectory and improve the accuracy of prediction model has become a key problem to be solved.In the existing location prediction research,researchers improve the accuracy of location prediction by integrating the spatiotemporal characteristics,location semantics and other factors in the location traj ectory.However,when these researches regard users' activity as location semantics,they lack considering the effect of sequence characteristics of user activity semantic on location movement patterns,so the accuracy improvement of location prediction is limited.In this paper,two location prediction models are proposed to solve the problem and to construct the sequence feature of user activity semantic based on cellular data.Firstly,ACP-RNN(Activity +Context+Profile+RNN)model is constructed based on the user activity sequence features extracted manually.This model verifies the influence of the user's activity sequence features on the movement mode.Combined with context features and user profiles,the predictive performance of the model is further improved.Secondly,this paper uses the recurrent structure in the Recurrent Neural Network to automatically extract the sequence features of user activity,and constructs a three-layer symmetrical recurrent neural network for the location prediction model TS-RNN(Three-layer Symmetrical Recurrent Neural Network).It maintains the stability of the model's location prediction performance and saves the workload of manual feature extraction at the same time.Because of TS-RNN's symmetrical model structure,it can not only predict next location based on the operator user's activity sequence,but also can predict next activity based on the users'traj ectory.TS-RNN model breaks through the limitations of existing prediction models for location prediction scenarios.In this paper,anonymous cellular data is used to analyze and verify the two location prediction models,and the performance advantages of the location prediction model based on activity sequence feature are proved.The final location prediction system not only improves the performance of location prediction,but also proposes a model scheme to save artificial feature extraction and extend the function of location prediction model.