Design And Implementation Of Mobile User Behavior Prediction Method Based On Big Data

Nowadays,increasing application of computers and networks appear,devices with position sensors are all running.They record the trajectories of moving users such as people and vehicles.The trajectory information expands with the time passed and the increase of their usage.So how to extract useful information from a large amount of trajectory information becomes a hot research field at present.A method of using trajectory information to predict user behavior was designed and implemented in this paper,including user trajectory prediction and user-specific behavior prediction.User trajectory prediction module can make use of a large amount of history user behavior trajectory information to predict a user's current activity trajectory in real time.This paper proposed a trajectory prediction method based on the improved SATP model.Firstly,for a large number of trajectory point data,we need to reduce them respectively according to the change of angle and MDL principle,thereby reducing the data to be processed to a certain extent.Then with a two-step clustering method,we cut back the number of states in the model.After that,we solved the problems that may exist in the original SATP model,such as staying or discontinuity of a state,by adding "self-transition" into this model.Finally,the problem that the prediction accuracy decreases because of the "over-simplification" in our method was solved by making the trajectory complete using the Bezier curve.In this paper,we also tested the effect of this method through experiments.By comparing the two-step clustering method with SinglePass and DBSCAN algorithm,it was found that the two-step clustering can greatly reduce the clustering time while maintaining the clustering effect using DBSCAN.At the same time,by comparing the improved SATP model with the original one,we found that the latter can significantly speed up the model training and prediction while only has a small accuracy decreasing,and ultimately meet the project requirements.Finally,we implemented and integrated this method into a module,and demonstrated the interfaces related to.In the the user specific behavior prediction module,this paper first designed its formal definition according to the complexity of a typical user specific behavior "pair behavior" in the project.Then,based on this definition,we proposed a method of identifying pair behavior using a large number of user trajectories,and used a grouping processing strategy to optimize this method.Based on the records of pair behaviors,we applied BP neural network and Naive Bayesian model respectively onto the prediction of different types of data designing the prediction method in detail.In this paper,we first tested the proposed method of pair behavior identification with different data volumes.We found that the time consuming increases with the increase of data volume,but generally meets the project requirements.At the same time,the interface implemented with Baidu map proved the correctness of this recognition method.In addition,this paper used different parameters to test the model respectively.Then we obtained the best model parameters through comparing the prediction errors.Using the best parameters in,the model is further tested in terms of training and predicting time consumption.The method proposed in this paper is effective and can be applied to this project.The two modules designed in this paper have been applied in the user behavior analysis system.