Data-Driven Car-Hailing User Value Analysis And Prediction

The rapid development of the online car-hailing in recent years has greatly solved the problem that passenger demand and supply services cannot be matched in real time.Passenger is the cornerstone of the online car-hailing system,their travel creates profit for the online car-hailing platform,which maintains the online car-hailing system.This paper focuses on online car-hailing passengers who have both consumption attributes and travel attributes.Their value and travel behavior are both crucial to the car-hailing platform.Online car-hailing passengers are large and various.Therefore,it is very necessary for the management of ride-hailing companies to study the user value and travel behavior of different groups.In this paper,a massive amount of online ridehailing order data and urban interest point data are used to study the value mining and value conversion prediction of online ride-hailing users travel and consumption.Firstly,by deeply mining the long-term,short-term and personalized value of online car-hailing users,the multi-dimensional value classification of users is carried out based on user portrait and clustering algorithm,and value distribution of different user groups is obtained.By means of distribution fitting,the differences of all kinds of users in each index are studied to further clarify the value connotation of users.Secondly,based on user value classification,the differences of users’ travel behaviors with different values are revealed through the variation trend of users’ hourly travel volume.The concept of travel topic is proposed to characterize the distribution of user travel.By mining the travel behaviors and topics of various users,a user travel topic model based on Dirichlet distribution is constructed to obtain a variety of typical travel topics of all online ride-hailing users and the travel topic distribution probability of different users.The travel topics of different s are significantly differentFinally,this paper focuses on the "temporal" change of user value,and adopts the method of combining clustering and classification to study the classification of user value by time period.Based on the features and labels in different time periods,the convolutional neural network(CNN)is used to enhance the input features,and the longshort-term memory neural network(LSTM)is used for "temporal" prediction.A CNN+LSTM-based method is proposed,which can predict the future value of online car-hailing users more accurately than traditional deep learning models.