Exploring the Patterns of Human Mobility Using Heterogeneous Traffic - PowerPoint PPT Presentation

Exploring the Patterns of Human Mobility Using Heterogeneous Traffic Trajectory Data Jinzhong Wang April 13, 2016 The UBD Group Mobile and Social Computing Laboratory School of Software, Dalian University of Technology

Outline Part 1 Introduction Part 2 Related Work Part 3 Methodologies Part 4 Experimental Results Part 5 Conclusion 1/24

Part 1 Introduction  In recent years, multi-source traffic data are more easily collected than before.  Analyzing and mining the laws of human mobility hidden in traffic big data have been a hot research field [1], [2].  There exist a variety of traffic trajectory data, which contributes to exploring the patterns of human mobility more deeply [3]. [1] J. Yuan, Y. Zheng, and X. Xie, “Discovering regions of different functions in a city using human mobility and POIs,” SIGKDD on Knowledge discovery and data mining. ACM, 2012, pp. 186–194. [2] Y. Zheng, L. Liu, L. Wang, and X. Xie, “Learning transportation mode from raw GPS data for geographic applications on the web,” WWW. ACM, 2008, pp. 247–256. [3] A. Noulas, S. Scellato, R. Lambiotte, M. Pontil, and C. Mascolo, “A tale of many cities: universal patter- ns in human urban mobility,” PloS one, vol. 7, no. 5, p. 37027, 2012. 2/24

Part 1 Introduction  To our knowledge, domestic and foreign researchers have proposed quite a few spatio-temporal patterns of human mobility. Power Law Exponential Law Log-Normal Law 3/24

Part 1 Introduction  They only focus on a single dataset which seems not to analyze the laws of human mobility very well.  The evaluation standards are different based on different studying data.  Subway smart transit data with 14 subway lines and Qiangsheng taxi data in Shanghai with 13 thousand taxis.  3 metrics (trip displacement, trip duration and trip interval).  Maximum likelihood estimation (MLE) and Bayesian information criterion (BIC). 4/24

Part 1 Introduction  We are the first to propose the distribution of human travel taking the subway.  We discover that the laws of trip displacement, trip duration and trip interval by subway and taxi. Trip Displacement Subway = Taxi Log-normal Trip Duration Subway != Taxi Weibull Log-normal Trip Interval Subway != Taxi Two-regime Weibull  We observe quite a few interesting phenomena, which contribute to inferring human mobility patterns more deeply by subway and taxi. 5/24

Part 2 Related Work  Processing Technology of Trajectory Data. • Substring tree to detect frequent movement sequences • Clustering algorithm to partition and group traces • Hidden Markov model to provide personalized recommendation • Principal component analysis to gain insight into the relation • A multivariate regression model to predict human mobility  Patterns of Human Mobility. Levy flight model • Power law distribution • Continuous-time random walk model • Radiation model • • Log-normal distribution Exponential law distribution • • Gamma distribution Power law with exponential cut-off • 6/24

Part 2 Related Work  The results are mainly based on single source trajectory data .  The patterns of human by subway are not investigated more deeply and thoroughly.  The statistical results lack a universal scaling law and are not independent of their own characteristics.  Therefore, our research is extremely indispensable and gains a valuable insight into human mobility patterns. 7/24

Part 3 Methodologies  We firstly collect the traffic data from Shanghai SODA (MLE BIC) competition website.  Then we leverage data manipulation language to clean the acquired datasets.  According to graph theory, we construct Human Mobility Network (HMN).  Especially, we introduce 3 evaluation metrics such as trip displacement, trip duration, and trip interval, and analyze two datasets quantitatively and comparatively.  Through using MLE and BIC, we finally explore the patterns of human mobility on weekend and weekday. Fig. 1. Overview of exploring the patterns of human mobility. 8/24

Part 3 Methodologies D s1 Subway weekend D s2 Subway weekday D t1 Taxi weekend D t2 Taxi weekday  In this paper, we use two datasets generated by subway smart card transactions (D s ) and Qiangsheng taxi GPS trajectory (D t ) from Shanghai in China.  Both (D s ) and (D t ) contain a total of more than 451 million trading records by 14 subway lines and 34 billion GPS records by about 13695 taxis respectively. 9/24

Part 3 Methodologies  Definition 1. A Human Mobility Network (HMN) G = (V,E) is a directed graph.  V denotes the set of the origins or the destinations.  For subway, V denotes the latitude and longitude coordinate of a subway station or picking-up and dropping-off location for taxi.  E represents the set of edges which connect Fig. 2. Human mobility network(HMN). an origin location to a destination. 10/24

Part 3 Methodologies  Definition 2. A Trip Displacement (d) is the spherical distance with a pair (v i ,v i+1 ) in HMN.  the spherical distance is the inferior arc length by using great-circle between v i and v i+1 on the surface of a sphere. r denotes the Earth radius with the value of 6370 km ϕ 1 and ϕ 2 represent the origin’s longitude and latitude respectively ψ 1 and ψ 2 represent the destination’s longitude and latitude separately. 11/24

Part 3 Methodologies  Definition 3. A Trip Duration (t d ) is the elapsed time between an origin v i and a destination v j in HMN. It is defined as follows: where t j and t i denote the drop-off time and the pick-up time of a trip.  Definition 4. A Trip Interval (t v ) is the elapsed time between two consecutive trips by the same person taking the subway or the same taxi. t ’ j denotes the drop-off time of the last trip. In a word, the travel time consists of t d and t v alternatively on human mobility. 12/24

Part 3 Methodologies  We introduce 4 widely used models to evaluate our 2 datasets, and unveil the best fitting distribution. The 4 distributions are showed in the following Equation4-7. Power Exponential Log-Normal Weibull 13/24

Part 3 Methodologies  We use MLE to estimate the parameters in every distribution. Then we continue to calculate the value of BIC as follows:  According to the BIC value of every distribution, we go a step further to obtain the Bayesian weights W i to determine which is the best law. 14/24

Part 4 Experimental Results (c) 17:00-19:00 on weekend (a) 7:00-9:00 on weekend (b) 11:00-13:00 on weekend 15/24 (d) 7:00-9:00 on weekday (f) 17:00-19:00 on weekday (e) 11:00-13:00 on weekday

Part 4 Experimental Results (c) 17:00-19:00 on weekend (a) 7:00-9:00 on weekend (b) 11:00-13:00 on weekend 16/24 (d) 7:00-9:00 on weekday (f) 17:00-19:00 on weekday (e) 11:00-13:00 on weekday

Part 4 Experimental Results  Trip Displacement. 17/24

Part 4 Experimental Results  Trip Displacement. 18/24

Part 4 Experimental Results  Trip Duration. 26 min Subway weekend 28 min Subway weekday 8min Taxi weekend 19/24 8min Taxi weekday

Part 4 Experimental Results  Trip Coefficient. 20/24

Part 4 Experimental Results  Trip Interval. 21/24

Part 4 Experimental Results  Trip Duration.  90% of trip interval by subway is within 580 min on weekend and 643 min on weekday respectively.  In other words, folks go to another consecutive trip earlier about 1 hour on weekend than weekday,  which implies that most individuals travel for social activity rather than working on weekend. 22/24

Part 5 Conclusion  In summary, we find that human mobility patterns in Shanghai have its own characteristics differing from existing statistical results.  Trip displacement is better fitted to the log-normal distribution rather than the exponential model.  The Weibull distribution fits the elapsed time by subway, whereas the log-normal distribution fits the trip time by taxi.  The patterns of trip interval by taxi follow the Weibull distribution, whereas there exist two-regime laws by subway which is the first part following the Weibull distribution and the tail part following the log-normal distribution. 23/24

THANK YOU FOR YOUR ATTENTION 24/24