The recent emergence of new forms of geo-social data, deriving from social media, sensors, and mobile phones, calls for an update to the methodological toolbox of social sciences. The new methods and tools need to harmonise with the inherent characteristics and challenges of the emerging data sources. This talk demonstrates how SocialGlass, a web-based system for (real-time) urban analytics, helps improve the understanding of human dynamics in modern-day cities, by capitalising on new geo-social data and pioneering data science techniques. Emphasis is on real-world applications, regarding social area analysis, crowd dynamics during large-scale events, and location prediction of new urban functions across different cities.
Presentation at the Centre for BOLD (Big, Open & Linked Data) Cities anniversary meet-up | Erasmus University Rotterdam -- May 29, 2017
Understanding and predicting urban dynamics through new forms of geo-social data: The SocialGlass system and its applications
1. Understanding and predicting
urban dynamics
through new forms of
geo-social data
The SocialGlass system
and its applications
Dr Achilleas Psyllidis
A.Psyllidis [at] tudelft [dot] nl
2. Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
00
Take-home message
Geo-social data are a goldmine of knowledge about cities
that we cannot afford ignoring
New forms of geo-social data call for an update to the
methodological toolbox of urban studies
to be continued…
3. Outline 01
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Geo-social data
New methods & tools
The SocialGlass system
Applications
Outlook
4. 02
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Geo-social data
Sources of geo-social data
• Census
• GPS
• Geo-portals,
Spatial data infrastructures
• Cell phones
• Location-based social networks
(e.g. Foursquare)
• Geo-enabled social media
(e.g. Twitter, Instagram etc.)
5. 03
Geo-social data
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Sources of geo-social data
• Census
• GPS
• Geo-portals,
Spatial data infrastructures
• Cell phones
• Location-based social networks
(e.g. Foursquare)
• Geo-enabled social media
(e.g. Twitter, Instagram etc.)
Traditional New
6. 04
Geo-social data
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
The past…
• Data scarcity
• Limited official resources (e.g. censuses, surveys)
• Large volumes, yet infrequently updated
• Limited storage and processing
• [+] Structured datasets
Spatial analysis methods were developed in this context…
7. 05
Geo-social data
…and the present
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
• Data richness
• Variety of sources
• Near real-timeupdates
• Abundant storage and processing
• [—] Spontaneous and unstructured datasets
…and are still in use
Need for an update to the methodological toolbox
8. 06
New methods & tools
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
An updated toolbox needs to capitalise on…
• High spatial & temporal resolution
• Ease of access (e.g. through APIs)
• Multiple information layers (e.g. spatial,temporal, social etc.)
9. 07
New methods & tools
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
…and tackle…
• Biases (representational, contextual, functional etc.)
• Complexity,diversity & multidimensionality
• Very large volumes
Human Computation
Machine Learning
Distributed Computing
How?
10.
11. 09
Anatomy of a tweet
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Age
Gender
Nationality
User info
Place of
residence
Date
Type of
activity
Place of activity (Location)
Social information
12. 10
The SocialGlass system
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
• Scalable web-based system*
• (Real-time)urban analytics & geo-visualisation**
• Used by IBM and the Municipalities of
Amsterdam, Paris, Adelaide
• 10+ Research Exhibitions & Demonstrations
(http://www.social-glass.org)*Bocconi et al.(2015) **Psyllidis et al. (2015a, b)
13. 11
The SocialGlass system
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
SocialGlass: choropleth map of prevalent social activities, as inferred from Instagram (http://www.social-glass.org)
14. 12
The SocialGlass system
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
SocialGlass: activity patterns of Amsterdam residents (ALF 2015) (http://www.social-glass.org)
15. 13
The SocialGlass system
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
SocialGlass: Flows of residents (http://www.social-glass.org)
16. 14
Applications
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Duration of experiment
Scale
SAIL 2015 ALF 2015
Regionalisation &
POI Location Prediction
1 Week 1 Month 3 Months
1 City
3 Cities
17. To what extent can a large-scale event influence
the activity patterns of different geo-demographic groups?
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
15
Amsterdam Light Festival 2015
18. 16
Amsterdam Light Festival 2015
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Residents’ activity appears balanced over time and dispersed across space throughout the monitoring period*.
*Psyllidis (2016)
19. 17
Amsterdam Light Festival 2015
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Tourists’ activity tends to cluster around central areas of the city and shows fluctuations over time*.
*Psyllidis (2016)
20. 18
Amsterdam Light Festival 2015
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
*Psyllidis (2016)
Local Moran’s I cluster mapsof social activity before, during, and after ALF
Moran’s Icluster map (normalised POIlocations)
Discrepancy between the clustering
of urban functions and the
clustering of activities*.
21. 19
Applications
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Duration of experiment
Scale
SAIL 2015 ALF 2015
Regionalisation &
POI Location Prediction
1 Week 1 Month 3 Months
1 City
3 Cities
22. 20
SAIL 2015
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
To what extent can social media microposts be used to
estimate the density of attendees in a large-scale event?
25. 23
SAIL 2015
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Visitors’ home countries Dutch residents’ homes
Crowd dashboard: Flow and route choice,
Visitors’ density (number of people / m²), Speed
and visit duration.
SocialGlass real-time analytics: Geo-location
mapping, Human mobility and dynamics, Visitors’
demographics.
Approach: Density estimation strategies inspired
by pedestrian flow theory
26. 24
Applications
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Duration of experiment
Scale
SAIL 2015 ALF 2015
Regionalisation &
POI Location Prediction
1 Week 1 Month 3 Months
1 City
3 Cities
27. 25
Regionalisation & POI location prediction
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
How could we detect neighborhoods of uniform social
interaction and predict new POI locations?
28. 26
Regionalisation & POI location prediction
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Age Gender Hour Topic
Social Category Venue Category Weekday All Clusters
Multidimensional clusters of social interaction in Amsterdam (Component planes and Hierarchical GeoSOM)*.
*Psyllidis et al. (2017)
29. 27
Regionalisation & POI location prediction
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Residents’ clusterLeisure cluster Tourists’ cluster
Arabic language clusterTransport / Nightlife cluster Work-related activities cluster
*Psyllidis et al. (2017)
30. 28
Regionalisation & POI location prediction
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
*Psyllidis et al. (2017)
Event spaceCafé Gym
RestaurantHotel Tram stop
Estimates of appropriate locations for various types of new POIs, based on a Factorisation Machine model*.
31. 29
Outlook
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
What’s next?
• SocialGlass as a service
• Future applications:
• Evidence-based policy for vulnerable youth (H2020, NWA)
• Data-driven policy for ageing populations in cities (H2020 – EU/China)
• Detection, prediction, and semantic enrichment of traffic incidents
32. 30
Take-home message (…continued)
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Geo-social data are a goldmine of knowledge about cities
that we cannot afford ignoring
New forms of geo-social data call for an update to the
methodological toolbox of urban studies
that combines machine learning and
spatial data science with human computation and
user modelling
33. 31
Relevant Publications
Achilleas PsyllidisUnderstanding and predicting urban dynamics through new forms of geo-social data
Psyllidis, A., Yang., J., Bozzon, A. (2017). Using MachineLearning on TwitterData to Regionalize Social Interactions and Predict New POI
Locations. PLoS ONE (underreview).
Psyllidis, A. (2016). RevisitingUrban Dynamics through Social UrbanData:Methods and tools fordata integration, visualization, and exploratory
analysis to understand the spatiotemporal dynamics of human activity in cities. PhDdissertation. A+BE|Architecture and the Built
Environment, Delft. doi: http://dx.doi.org/10.7480/abe.2016.18
Psyllidis, A., Bozzon, A., Bocconi, S., & Bolivar, C. T. (2015a). Harnessing Heterogeneous Social Datato Explore, Monitor, and Visualize Urban
Dynamics. In: Ferreira J Jr, Goodspeed R (eds) Planning Support Systems and Smart Cities: Proceedings of the 14th International
Conference on Computers in Urban Planning and Urban Management (CUPUM 2015). MIT, Cambridgre, MA, USA,pp. 239-1 — 239-22.
Psyllidis, A., Bozzon, A., Bocconi, S., & Bolivar, C. T. (2015b). A Platform forUrban Analytics and SemanticIntegration in City Planning. In: Celani
G, Moreno Sperling D, Franco JMS (eds)Computer-Aided Architectural Design Futures – NewTechnologies and the Future of the Built
Environment: 16th International Conference(CAADFutures 2015) – Selected Papers. LNCS, CCIS 527, Springer, Berlin Heidelberg, pp.
21—36. doi: http://dx.doi.org/10.1007/978-3-662-47386-3_2
Bocconi, S., Bozzon, A., Psyllidis, A., & Bolivar, C. T. (2015). SocialGlass: A Platform forUrban Analytics and Decision-making Through
Heterogeneous Social Data. In: Gangemi A, Leonardi S, Panconesi A (eds) 24th International Word Wide Web Conference (WWW 2015).
ACM, New York, NY, pp. 175—178. doi: http://dx.doi.org/10.1145/2740908.2742826