Research On Hierarchical Taxi Pick-Up Recommendation

Taxi pick-up recommendation service is an important part of intelligent transportation,which can not only improve the efficiency of taxi occupancy,but also reduce environmental pollution and traffic congestion.In recent years,with the continuous development of Internet of Things perception and high performance computing technology,taxi trajectory data preservation and acquisition become more convenient,which provides a cornerstone for the development of taxi pick-up recommendation services.How to use data mining and other means to identify and discover the pattern of taxi pick-up behavior information is the key means of taxi pick-up recommendation.This paper mainly uses taxi trajectory data as the research object,has carried on the taxi pick-up recommendation correlation research.The main research contents include:1.Taxi pick-up area recommendation based on the short-term GPS data can greatly reduce the overhead and improve the efficiency,but it is challenging to deal with the problems of data sparseness.In recent years,some scholars have proposed a recommended taxi pick-up area method based on matrix decomposition to deal with data sparsity.However,the accuracy of recommendation needs to be further improved because the interaction between features is not considered.Therefore,we propose a recommendation model of taxi pick-up area based on factorization machine.By introducing hidden vectors,the factorization machine can estimate the characteristic parameters,which can adapt to the highly sparse data recommended system environment.In this paper,we mine more recommended features from the trajectory data,which are used to accurately describe the spatiotemporal characteristics of the area and the driver’s personalized preference.A general factorization machine is used to predict the driver’s access probability to the area as the basis for the pick-up area recommendation.The experimental results show that the proposed method is superior to the traditional taxi pick-up area recommended algorithm in prediction accuracy.2.In order to further improve the recommendation performance and learn complex feature interaction,we propose a new method,which combines the Deep FM and the spatio-temporal context to recommend the taxi pick-up area.First,the study area is divided into several grids with equal size,and map both pick-up points and POI to the corresponding grids.Then,the spatiotemporal context corresponding to the pick-up area is extracted from the pick-up point and POI.Finally,these spatio-temporal contexts are fused into Deep FM to learn complex feature interaction.Experiments show that our approach can significantly improve the recommendation accuracy and is superior to some state-of-the-art methods.3.Taxi hierarchical pick-up recommendation method combining offline and online is proposed.The system uses the taxi pick-up area model that integrates Deep FM and temporal context as the offline part.It uses the generated driver-time slot-grid access probability matrix and combines the spatiotemporal information of the current no-load driver to obtain the recommended area.The online recommendation carries out spatial-temporal analysis on the recommended area,obtain the candidate pick-up point which can represent the passenger gathering place,and measure the attribute of the candidate pick-up point,so as to provide the driver with fast and accurate passenger point recommendation.The purpose of this paper is to improve the carrying passenger efficiency of no-load taxi and overcome the problem of data sparseness.Therefore,a recommendation model of taxi pickup area based on factorization machine is proposed.In view of the problems existing in this model and in order to further improve the recommendation performance,a taxi pick-up area recommendation method fusing the deep factorization machine and the spatio-temporal context is proposed.On the basis of pick-up area recommendation,in order to provide high precision and good time performance to drivers,we put forward the Taxi hierarchical pick-up recommendation method combining offline and online.