Exploring Human Mobility with Multi- Source Data at Extremely Large Metropolitan Scales Authors: Zhang, Zhang, Huang, Xu, Li, He University of Minnesota, USA Chinese Academy of Sciences, Shenzhen, China presented by: Niklas Lensing
Outline Motivation Architecture of the solution Solution Evaluation Comment Summary and future work
Motivation Why would one like to track human mobility in metropolitan areas?
Motivation author‘s answers: improve wireless services in urban areas improve urban efficiency i.e. public transport, traffic management
Motivation Assuming you want to track mobility of people, which technologies could you use?
Motivation several ways to track mobility why only use a single data source? the first ones to use multi-source data Taxis Subway Buses phone call records track human mobility more precisely
Architecture of the solution
Solution How to get significant mobility data from the raw data?
Solution separate Shenzhen into 496 regions
Solution two types of data: transit data: subway, bus, taxi cellphone data cellphone data are more precise than transit data cover more regions
Solution there are less cellphone data than transit data cellphone data are only considered when at least two cell towers where used authors found out that cellphone data have a highly repeatable pattern to outweigh the effectiveness of transit data historical cellphone data has to be used problem: processing time and storage
Solution how to only use as much historical cellphone data as needed? effectiveness of cellphone data on one day varies between region pairs cellphone data < transit data use more historical cellphone data cellphone data > transit data use less historical cellphone data
Solution cellphone data is more precise than transit data to make cellphone data more effective than transit data, historical data needs to be used transit data to cellphone data ratio indicates how much historical data is needed functionality of the Mobility Abstraction layer
Evaluation compared with two state of the art mobility models: Radiation (transit data) WHERE (cellphone data) used a very precise dataset as „ground truth“ mPat scored best in every test WHERE scores well on long time data records Why not using „ground truth“ dataset?
Evaluation real world experiment: Inter Region Transit (IRT) find region pairs with high human mobility but low transit data establish point to point connections travel time reduced by 36% compared to public transport
Comment repeatable pattern in cellphone data true in general? What if call starts/ends too early? Are private cars covered? privacy
Summary and future work first multi-source human mobility model outperforms state of the art human mobility models in future will be included: bike rental onboard GPS of private cars enforce privacy to motivate residents‘ participation
谢谢! Thank you!