Recommending a personalized sequence of pick-up points
Introduction
The advances in location-acquisition, wireless communication, and mobile computing techniques have enabled us to generate a group of location-based services. Taxis, frequently traveling in the city every day, have become an indispensable part of the intelligent transportation system. But taxis spend 35%–60% working time to search passengers by cruising along the roads [22], [15].
To effectively reduce tail gas pollution and improve drivers' profits, many researchers mostly focus on pick-up point recommendation based on the path-optimization or the profit-maximization strategy. Then the nearest or the most valuable point is chosen for taxis. Obviously, there are some defects: (1) It may result in sending all cruising taxis to the same location to compete for the same group of passengers. (2) The historical pick-up points are lack of in-depth analysis.
To address above problems, we propose a personalized sequence of pick-up point recommendation (PSPPR) approach. It aims to provide customized sequence recommendation service by mining taxi-drivers’ preferences from spatial-temporal analysis on historical pick-up points. Our contribution lies in three aspects:
- (1)
To improve the recommendation accuracy of single point, we present a novel model of spatio-temporal analysis (STA). It consists of four algorithms: first space last time analysis (FSLTA), First time last space analysis (FTLSA), improved The OPTICS algorithm, and candidate-points filtering via voting mechanism. Experimental results show that our STA model achieves satisfactory performance.
- (2)
To meet the customized need for the individual taxis, we combine a suitable passengers-finding strategy with a personalized sequence recommendation. Firstly, we analyze drivers’ preferences by categorizing their historical pick-up points in light of the nearest points-of-interest (POI). Secondly, we put forward a probability optimization model (POM) in which passengers-finding probabilities are calculated to create candidate-points and these points are filtered in turn to generate a personalized sequence.
- (3)
Our PSPPR system achieves satisfied performance. Drivers’ preferences are inferred from POIs’ attributes. Moreover, we combine item-based collaborative filtering with POM to recommend a personalized sequence of pick-up points.
The remainder of this paper is organized as follows. Section 2 describes the related works. The framework of our PSPPR approach is presented in Section 3. Next, we will illustrate the spatial-temporal analysis model in Section 4, the reference analysis method in Section 5, and the probability optimization model in Section 6. Futhermore, Section 7 gives our experimental results. And conclusions are drawn in Section 8.
Section snippets
Related work
The advance in location acquisition technologies has generated a myriad of spatial trajectories, which offer us fostering a broad range of applications in urban computing [28], vehicular networking [29], [30], [31], intelligent transportation systems [36], [37], route prediction [38], travel recommendation [39], [40], and mobile crowd photographing [41], [42], etc. The prevalence of these applications in turn calls for systematic research, such as trajectory data mining [27], spatio-temporal
The framework of our PSPPR approach
Pick-up points are the locations where taxi-drivers find passengers. Its distribution can reflect drivers’ behaviors to find passengers. If we discover the behavior rules of taxi-drivers, we could apply it to intelligent transport, urban planning, traffic management, etc.
The framework of our approach is given in Fig. 1. In the framework, some novel methods, such as spatio-temporal analysis model, probability optimization model, and personalized sequence recommendation, are proposed.
First of
Spatial-temporal analysis (STA)
In this section, we present the details of spatio-temporal analysis, which generate the candidate points, describe how we decide the type of these points, and compute the passenger finding probability in every point.
Preference analysis
Taxi-passengers’ occurrence exists some regularities in the spatio-temporal distribution. The regularities are observed from different angles and then are reflected on drivers’ personalized behaviors. Perference analysis is proved to be valuable for related applications, such as travel package recommendation [40].
Pick-up points within a neighborhood always contain some POIs, such as a restaurant, a hospital, a station, etc. According to the functional classes that these POIs are, the regional
Probability optimization model
Each taxi-driver has his/her driving style and driving habits. The decisions they have made are definitely relative to their preferences and experiences, such as the features of cruising roads and the types of pick-up points. Therefore, it should be a kind of attentive services to recommend personalized pick-up points for different drivers by mining their preferences from historical GPS data.
Experiments
Our data of experiments is public trajectory data set provided by Microsoft Research Asia, generated by GeoLife project. It contains 182 taxicabs’ previous trajectory from April 2007 to August 2012 in Beijing. The data of POI come from www.datatang.com.
Our method is consisted of candidate-points generation and recommending a sequence of candidate-points. Firstly, we evaluate the accuracy of the generation of candidate points. Next, we compare the results of our approach with the classical Top-k
Conclusions
The advances in location-acquisition and mobile computing techniques have generated massive spatial trajectory data of a diversity of moving objects, such as people, vehicles and animals. The latent values of spatial trajectory data have generated a group of location-based services. Pick-up points recommendation by mining taxis’ trajectory can effectively improve drivers’ profits and reduce consumption. However, the recommendation accuracy of existing approaches is not good enough owing to the
Acknowledgements
This research has been supported by National Nature Science Foundation of China (61572187, 61370227, 61572186), China National Key Technology R&D Program (2015BAF32B01), Hunan Provincial Natural Science Foundation of China (2015JJ2056), Hunan Provincial University Innovation Platform Open Fund Project of China (14K037), General Project of Hunan Provincial Education Department (16C0642).
Yizhi Liu received the Ph.D. degree in Computer Application Technology at the Institute of Computing Technology, Chinese Academy of Sciences, China, in 2011. He is now an associate professor and works for School of Science and Engineering, Hunan University of Science and Technology, China. His current research interests are in the fields of multimedia content analysis, trajectory mining and location-based services.
References (39)
- et al.
An effective taxi recommender system based on a spatio-Temporal factor analysis model
Inf. Sci.
(2015) - et al.
Understanding taxi service strategies from taxi GPS traces
IEEE Trans. Intell. Transp. Syst.
(2015) - et al.
Fusing audio vocabulary with visual features for pornographic video detection
Elsevier: Future Gener. Comput. Syst.
(2014) - et al.
Extracting salient region for pornographic image detection
Elsevier: J. Vis. Commun. Image Represent.
(July 2014) - et al.
Offloading mobile data traffic for QoS-aware service provision in vehicular cyber-physical systems
Future Gener. Comput. Syst.
(2016) - et al.
Robust Common Visual Pattern Discovery Using Graph Matching
Visual Commun. Image Represent.
(2013) - et al.
Context-Aware taxi demand hotspots prediction
Int. J. Bus. Intell. Data Mining
(2010) - et al.
Hunting or waiting? discovering passenger-finding strategies from a large-scale real-world taxi dataset
Proceedings of the 8th IEEE International Workshop on Managing Ubiquitous Communications and Services
(2011) - et al.
T-Finder: a recommender system for finding passengers and vacant taxis
IEEE Trans. Knowl. Data Eng.
(2013) - et al.
Recommending pick-up points for taxi-drivers based on spatio-temporal clustering
Proceedings of the 2nd IEEE International Conference on Cloud and Green Computing
(2012)
Where to find my next passenger
Proceedings of the 13th ACM International Conference on Ubiquitous Computing
An energy-efficient mobile recommender system
Pro-ceedings of the 16th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
Towards Efficient Vacant Taxis Cruising Guidance
Pro-ceedings of the IEEE Global Communications Conference
Locating Lucrative Passengers for Taxicab Drivers
Proceedings of the 21 st ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems
HUNTS: A Trajectory Recommendation System for Effective and Efficient Hunting of Taxi Passengers
Proceedings of the 14th IEEE International Con-ference on Mobile Data Management
Backward path growth for efficient mobile sequential recommendation
IEEE Trans. Knowl. Data Eng.
Towards reducing taxicab cruising time using spatio-temporal profitability maps advances in spatial and temporal databases
LNCS
Pick-up tree based route recommendation from taxi tra-jectories web-Age information management
LNCS
P-Cruise: reducing cruising miles for taxicab networks
Proceedings of the 33rd IEEE Real-Time Systems Symposium
Cited by (15)
AdaBoost-Bagging deep inverse reinforcement learning for autonomous taxi cruising route and speed planning
2023, Transportation Research Part E: Logistics and Transportation ReviewAdaptive passenger-finding recommendation system for taxi drivers with load balancing problem
2022, Computers and Industrial EngineeringCitation Excerpt :However, this approach is too strict and inefficient since the number of taxis is different from the number of anticipated passengers all the time. Liu et al. (2018) and Wang et al. (2018) proposed personalised passenger-finding recommendation methods that consider driver preference, i.e. degree of familiarity for a specific area. Since the recommendation procedure is personalised, these methods can prevent sending the same recommendation to multiple drivers.
Research on the big data of traditional taxi and online car-hailing: A systematic review
2021, Journal of Traffic and Transportation Engineering (English Edition)Citation Excerpt :This helps reduce taxi drivers' searching time, road congestion, and air pollution emission. It also benefits the effective increase of drivers' profits (Liu et al., 2018; Zhang et al., 2012) and passenger's satisfaction (Hu et al., 2014), and helps to decide the overall layout of taxi service locations and driving route planning (Zong et al., 2019). We first summarized and divided the strategy of taxi driver-searches into drivers' autopsychic and platform dispatching respectively.
Trajectory analysis for on-demand services: A survey focusing on spatial-temporal demand and supply patterns
2019, Transportation Research Part C: Emerging TechnologiesCitation Excerpt :This might cause congestion. To solve this problem, Liu et al. (2018) proposed a method for recommending personalized parking places. They added a filter process to guarantee that the recommended parking places and the drivers' historical pick-up locations were similar in the pattern of point of interest (POI).
Hot spots areas mining of online ride-hailing based on D-OPTICS algorithm
2023, Beijing Hangkong Hangtian Daxue Xuebao/Journal of Beijing University of Aeronautics and AstronauticsFusing Spatio-Temporal Contexts into DeepFM for Taxi Pick-Up Area Recommendation
2023, Computer Systems Science and Engineering
Yizhi Liu received the Ph.D. degree in Computer Application Technology at the Institute of Computing Technology, Chinese Academy of Sciences, China, in 2011. He is now an associate professor and works for School of Science and Engineering, Hunan University of Science and Technology, China. His current research interests are in the fields of multimedia content analysis, trajectory mining and location-based services.