Geographic data mining of online social networks

Geographic data mining of online social networks , GIS Research UK (GISRUK) Durham University 3/4/2009

Transcript

1. GEOGRAPHIC DATA MINING OF ONLINE SOCIAL NETWORKS Alex D Singleton Department of Geography and Centre for Advance Spatial Analysis , University College London
2. SOCIAL NETWORKS • Method of expressing the social environment as “patterns of regularity among interacting units” (Wasserman, 1994) • Social actors range in scale and form - individuals, areas, companies ... • Relational data - link actors together
3. Crossley, 2009: Post Punk, Manchester
4. Tampubolon, 2006: FTSE 100 Executives Connections by Non-Executive Directors
5. ARTICLE IN PRESS Fig. 1. Illustration of the hotel network as generated by Pajek. Hotels are indicated with square symbols; people with circle symbols. Each line connects an IDU with a specific hotel, as named by them as a site where they had injected drugs in the past 6 months. The hotels numbered 1–11 are the 11 hotels identified as 3-core. 3-core IDU are indicated by K; 2-core IDU as ; and 1-core IDU as J. J.L. Wylie et al. / Health & Place 13 (2007) 617–628 621 Wylie, 2006: Injection Drug Use and Geographic Setting (Hotels)
6. LANGUAGE OF SN Node (Vertices) Lines (Edges) Binary (1,0) Directional (+,-) Continuous Integer
7. SN AND SOCIAL CAPITAL • “features of social life - networks, norms and trust - that enable participants to act together more effectively to pursue shared objectives” (Putnam, 1995:664) • Individual V Group or Area • How does this translate online?
9. SOCIAL NETWORK ONLINE • Facebook (www.facebook.com) • May 2008 - 123.9 million unique users (McCarthy, 2008) • API released 2006 - 650,000 developers world wide
10. Touchgraph Clusters - Based on friend links What about node characteristics?
11. • June 2006 • GB surnames search • Ethnicity statistics • Social status • Experian Mosaic NAMES DATA
12. NAME SOCIAL STATUS • Modified electoral roll • Address Level (postcode) • Geodemographic - Acorn, CACI • 5 Categories, 17 Groups, 56 Types • Rank Types by wealth
13. NAME SOCIAL STATUS • Append wealth rank to forename and surname by postcode • Create median, and standard deviation rank for each name, rank name by median, then lowest standard deviation where ties • Person scores created by joining forename and surname scores • Name ranks used to create a measure at OA This is not very scientific - just for fun!
14. Jeremy Paxman 99 Judi Dench 94 Sebastian Coe 88 Christian Bale 78 Nigella Lawson 74 Cliff Richard 69 Rio Ferdinand 62 Elton John 57 Robert Carlyle 52 Robbie Williams 49 David Beckham 42 Martine McCutcheon 39 John Lennon 33 Terry Christian 30 Jim Carrey 23 Britney Spears 17 Coleen McLoughlin 7 Wayne Rooney 5
15. PHP API UCL Server MYSQL Client Browser
16. MYNAMEPROFILER • http://apps.facebook.com/mynameprofiler You Friends Area Friends Ranked
17. User Install P Friends Friends of Friends P P P P P P P P P P
18. © Crown copyright 1999 0 50 100 km • How do virtual connections map onto real world locations? • Places, geodemographics • How are virtual connections created by real world associations? • Edge V Node characteristics • What do these associations tell us about social capital?
19. • Launch of the site - probably press release to BBC • Data • Modified community detection algorithm: examine both edges and nodes • Examine the geodemographic profiles of those clusters which are identified NEXT STEPS