Understanding Human Mobility. Mirco Nanni. Knowledge Discovery and Data Mining Lab (ISTI-CNR & Univ. Pisa) kdd.isti.cnr.

Transcription

1 Understanding uman Mobility Mirco Nanni Knowledge Discovery and Data Mining Lab (ISTI-CNR & Univ. Pisa) kdd.isti.cnr.it

2 Mobility: Objectives Infer human mobility from mobility proxies GPS traces Mobile phone records Social media networks Twitter, Flickr, extract mobility models and patterns Regularities routines, traffic distribution, Predictive models apply them to specific services Towards mobility planners / managers Towards citizens, etc.

3 (A basic) Urban Mobility Atlas Vehicle mobility from GPS traces

4 Analysis Methods Main general line: infer semantics of places and movements from raw mobility data Methods vary with data sources GPS trajectories Mobility profiles Individual Mobility Network GSM call records Routines discovery Sociometer Others: e.g. PT ticket validation data

5 Methods.GPS.Mobility Profiles Describe an abstraction in space and time of the systematic movements of a user. The exceptional movements are completely ignored. Individual istory Trajectory Clusters Routines Routines = trips that most likely will take place also in the future Applications in prediction and carpooling

6 Methods.GPS.IMN Individual Mobility Networks Extract relevant locations and their trips

7 Methods.GPS.IMN Target: guess activities for locations and trips Method: learn a sequential classifier from labelled IMNs over network features

8 (A better) Urban Mobility Atlas Show collectively frequent paths Systematic vs. occasional mobility

9 Application 1 GPS.Proactive Carpooling Routine containment: driver r2 can give a lift to driver r1 Carpooling Network: who might share a ride with whom Carpooling Assignment: optimal matching to maximize impact N N'

10 Application 2 GPS.Trajectory Prediction Combination of individual habits (profiles) and collective common behaviours

11 Methods.GSM.Routines A single trace of an individual can be poorly informative about his/her movements B W A C A B W C time

12 Methods.GSM.Routines Yet, several daily traces of the same individual might allow to identify regular places and trips W A C W B W A W A B W W C

13 Methods.GSM.Routines The whole individual mobility is then summarized by its systematic movements Afternoon routine W Morning routine They will be used as typical daily schedule of the individual

14 Methods.GSM.Routines For each user, extract significant L1 L2 Aggregate (individual) systematic movements into (collective) systematic flows Examples: Outgoing traffic Incoming traffic

15 D4D Data challenge, 2013 Simulated traffic in Abidjan, Ivory Coast

16 Methods.GSM.Sociometer Derive presence distribution for each < user, municipality > t1 = [00:00-08:00) t2 = [8:00-19:00) t3 = [19:00-24:00) Cell12 24/06/ : Cell12 24/06/ : Cell15 25/06/ : Cell15 25/06/ : Cell11 25/06/ : Cell12 26/06/ :01.

17 Methods.GSM.Sociometer Recognize user's role through his fingerprints Based on learning a fingerprints classifier Resident A Visitor B A Commuter B A

18 Methods.GSM.Sociometer Sample application: measuring exceptional events resence of Visitors GSM - Pisa - istorical Center Presenze Internet Festival Presenze Presenza media settimanale Giovedi Venerdi Sabato Domenica Presence of Visitors GSM - Lucca Comics 2012 Presenze Comics 2012 Presenze Presenza media settimanale Giovedi Venerdi Sabato Domenica

19 Other methods GSM.Correlation Patterns GSM data for event detection in space & time +35% 8: : :00 Extract correlated groups/sequences of events +15 % +35% +5% +10 % {(Cell27,+35%)} {(Cell7,+15%),(Cell5,+10%)} {(Cell13,+5%)}

20 The future: multidimensional data Data Integration at the individual level Integrate GPS, GSM, Flickr, Twitter Personal Data Store view Controlled by the user Releases only essential information for services Data Integration at the collective level Comparative analysis of phenomena from different data sources E.g. event discovery & understanding w/ tweets & GSM

21 Novel well being indicators Dino Pedreschi Knowledge Discovery and Data Mining Lab (ISTI-CNR & Univ. Pisa) kdd.isti.cnr.it

22 Individual Mobility Networks Individual Mobility Data Store Semantic Amplifier for Big Data Individual vs Collective modeling S. Rinvizillo, L. Gabrielli, M. Nanni, L. Pappalardo, D. Pedreschi, F. Giannotti. The Purpose of Motion: Learning Activities from Individual Mobility Networks. In 2014 International Conference on Data Science and Advanced Analytics (DSAA 2014), 2014

23 eterogeneity Model the characteristic distance traveled by individuals the radius computed on the k most visited locations L. Pappalardo, F. Simini, S. Rinzivillo, D. Pedreschi, F. Giannotti, A.L. Barabasi, Returners and Explorers: Dichotomy in uman Mobility, Submitted to Nature Communications

24 Well-being Indicators Mobility Diversity Social Diversity