Research On Visual Inference Of User Behaviour Based On Mobile Call Data

According to the Global Mobile Market Report 2021 survey,there are 3.9 billion mobile phone users worldwide in 2021.The large number of mobile phone users generates large-scale communication data;mainly including network access data,Call Detail Record(CDR),etc.The CDR data contains temporal and spatial information about the subscriber.This provides the basis for studying the mobile behaviour of citizens in cities.However,CDR data is characterised by its large size,long time span and discrete nature,making it difficult to study and analyse directly.Visual analytics can map the data into various graphs and charts,provide a clear understanding of the data through visual channels,and allow for exploratory analysis through interactive methods.Techniques such as Natural Language Processing and Deep Learning can mine and predict the data to discover its underlying information.This paper therefore combines natural language processing,visual analytics and deep learning techniques to extract user behaviour from CDR data,combine visual analytics to analyse and infer user roles,and use deep learning to predict user trajectories.The main work of this paper is as follows.(1)A sequential base station trajectory embedding model and a user dimension reduction method are proposed.Social role refers to a set of individual behavior patterns associated with position in social life.Users' social role can be explored through the study of user behavior patterns.However,because of the large scale of call data and the large number of users,it is difficult to capture the behavior patterns of users.In view of these problems,this paper uses the base station sequence of the user's call to construct the user's trajectory sequence,and discovers the user's behavior pattern through the long-term trajectory sequence.Users with the same role will show similar trajectory sequence.Based on the method of Word embedding and Position embedding sequence data modeling,pos-cell2 VEC embedding model considering sequence order is proposed to recognize the semantics of the base station.Then,a method of user dimension reduction based on trajectory sequence embedding is proposed.The results of user dimension reduction can be visualized in a real sense by means of high-dimensional visualization,and similar users will gather together in low-dimensional space.(2)A user role visual analysis system was designed.This paper combines the PosCell2 Vec model to propose a visual analysis model for user role projection,designing interaction mechanisms such as scaling,dynamic clustering and temporal filtering,and designing and implementing a visual analysis system for user social role projection based on massive call data based on multi-view collaborative visual analysis technology.The system includes clustering diagrams,user base station access lists,user behaviour Gantt charts,group user spatio-temporal feature matrices and trajectory heat maps.The results of the case study with real data in this paper show that the analyst can effectively combine user state sequences and base station semantic information to infer user social roles(students,drivers and marketers)through iterative interaction with the system.(3)A user trajectory prediction method based on CDR data is proposed.It addresses the problems of large scale,sparsity and difficulty in analyzing and predicting the trajectory of CDR data.This paper constructs an Adaptive Trajectory Prediction Network(A-TPN)based on the Bi-Directional Long Short-Term Memory(BILSTM)model of deep learning,combining attention mechanism,user's contextual features and gating mechanism.Network);and conducts prediction analysis on users' trajectories.Experiments show that the prediction accuracy of A-TPN reaches 94.76% on real data sets,which is 8.7% better than the traditional Markov statistical calculation method.