Design And Implementation Of Telecom User Behavior Prediction System Based On Recurrent 3D Convolutional Neural Networks

Telecom network traffic is growing rapidly,and the need for real-time network traffic scheduling is becoming more important.At present,considerable progress has been made in network-area-granularity and day-level traffic prediction,but there has been a lack of prediction for user-granularity and day-level traffic and more other telecom behaviors.On the other hand,A large number of telecom packages have emerged to meet the differentiated needs of users.The personalized recommendation for the telecom package has already solved the problem of matching between the monthly package and the user,but still cannot perceive the user-granularity and day-level change of telecom user behavior,making it difficult to recommend the appropriate daily packages to users actively and timely.The above questions can be unified as how to provide a telecom user behavior prediction based on user granularity and day-level.Faced this problem,this paper proposes a 3D feature frame data structure,which abstracts the user-granularity and day-level telecom user behavior features under the premise of preserving characteristics of time series and local correlation.It proposes a Recurrent 3D Convolutional Neural Networks model,which well integrates the advantages of 3D Convolutional Neural Networks and Stacked Long-term and Short-term Memory Networks,better mining the behavior feature and learning the historical behavior rules of telecom users,thereby achiving automatically predicting the future behavior of telecom users.Firstly,this paper studies technique about telecom user behavior prediction,proposes the Recurrent 3D Convolutional Neural Networks model,and tunes it.Then,this paper analyzes the characteristics of telecom user behavior prediction problem,and clarifies the functional and non-functional demand to be met by the telecom user behavior prediction system.Then,according to the demand analysis,around the 3D feature frames and the Recurrent 3D Convolutional Neural Networks model,this paper describes the functional structure design of the overall telecom user behavior system and each module,and design and implement a complete telecom user behavior prediction prototype system based orn the MapReduce programming model and the TensorFlow programming framework.Finally,this paper describes testing for the prototype system,verifying the feasibility of the prototype system.The paper provides some significance for the future study of telecom user behavior prediction.