Mobility Trajectory Mining Human Mobility Modeling at Metropolitan Scales Sibren Isaacman 2012 Mobisys Jie Feng 2016 THU FIBLab
Syllabus Motivation Challenge and Contribution Data Description Model Spatial and temporal parameters Evaluation Application Discussion
Motivation CORE Previous work: WHERE: Goals: ad hoc, university campus, universal model Still need a realistic model: matches empirical observations for large and distinct geographic areas. WHERE: how large populations move within different metropolitan areas Goals: Individual: how they move between important places in their lives Aggregate: reproduce human densities over time at metropolitan areas Application: evaluate what-if scenarios Look at, not realistic, limited, not fit to specific area New Model How the addition of a new residential or employment area might change traffic patterns. CORE
Challenge and Contribution No semantic info: important place Coarse granularity: cell tower, active Contribution: synthetic CDRs: Less storage, open data, what-if Model: generate synthetic CDRs Validation and Application Large-scale validation: NY and LA Daily ranges for travel Overcoming challenge
Data Description CDRs US Census Data(detailed info) Home/work locations and commute distances Previously Published data Calling Patterns Enough No such data Performance loss [1] S. Almeida, J. Queijo, and L. Correia. Spatial and temporal traffic distribution models for gsm. In Vehicular Technology Conference, Sept. 1999. [5] J. Candia, M. C. González, P. Wang, T. Schoenharl, G.Madey, and A.-L. Barabási. Uncovering individual and collective human dynamics from mobile phone records. MATH.THEOR., 41:224015, 2008.
Model: Individual Detail Input: real CDRs/Census 60% Input: real CDRs/Census Step1:Home distribution Step2:Commute Distance Step3:Work …… Step1: Identify the key properties of human movement Step2: Use PDs to generate synthetic CDRS
Spatial and Temporal Parameters Spatial Information: Important Locations Previous work+: statistics and regression Spatiotemporal Information: Hourly Population Densities CDRs: call records = population Census: 7pm-7am(home),7am-7pm(work) Temporal Information: Calling Patterns PerUserCallsPerDay: average and std When Standard deviation of calls + S. Isaacman, R. Becker, R. Cáceres, S. Kobourov,M. Martonosi, J. Rowland, and A.Varshavsky. Identifying important places in people’s lives from cellular network data. In 9th International Conf. on Pervasive Computing, 2011.
Evaluation Earth Mover’s Distance (EMD) Comparison Models Synthetic pop density VS real pop density(real CDRs) Image test* Human readable Comparison Models Random Waypoint: RWP r ori->r des-> r vel->r wait -> …… r:random Weighted Random Waypoint: WRWP Selected home/work r velocity and r wait location shift transform EMD attempts to find the minimum amount of energy required to transform one probability distribution into another * Y. Rubner, C. Tomasi, and L. J. Guibas. A metric for distributions with applications to image databases. In Proc.IEEE International Conference on Computer Vision, 1998.
Application(boxplot) Daily ranges of travel Message Propagation Hypothetical Cities If 10% work at home
Discussion Group mobility Related work Future work? Model group migration between street blocks during the day Data: Mobile phone data Every group size is the point Related work Gravity model between cities Origin-Destination Matrix SIR model (susceptible, infected, removed) Individual model Ad-hoc network Future work? Community detection + OD matrix + something……