1 / 17

Camera Brand Congruence in the Flickr Social Graph

Camera Brand Congruence in the Flickr Social Graph. Adish Singla * , Ingmar Weber Ecole Polytechnique Fédérale de Lausanne (EPFL) * Now Microsoft Live Search. The main research question addressed:. If I use a Nikon camera, are my friends more likely to use a Nikon camera as well?. Sony.

june
Télécharger la présentation

Camera Brand Congruence in the Flickr Social Graph

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Camera Brand Congruence in the Flickr Social Graph AdishSingla*, Ingmar Weber Ecole Polytechnique Fédérale de Lausanne (EPFL) *Now Microsoft Live Search

  2. The main research question addressed: If I use a Nikon camera, are my friends more likely to use a Nikon camera as well? Sony Canon Does it depend on • whether we are in the same country? • whether we are close friends? • whether I use a cheap/expensive camera? • …. Canon Sony Relevant for advertising in social networks.

  3. We use data from Flickr ... Camera model/brand Date taken

  4. User‘s location User‘s contacts User‘s groups

  5. Extracted Information • Per-image • Camera brand • Camera model • Date taken • Per-user • Location • List of contacts • List of groups

  6. Data Pre-Processing • Map camera brand to ID • E.g. Minolta = Konika = Konica • Map camera model to ID • E.g. Maxxum 7D = Dynax 7D • Map location to country ID • E.g. California = Canada’s neighbor = USA • Get unique camera brand for users and “buckets” • March-May 2006, March-May 2007, March-May 2008 • Majority voting of (up to) 10 images in a bucket Get implicit brand information.

  7. Data Statistics • A complete connected component • 3.9M users, 67M edges (in summer 2008) • 1.2M users with brand information • 37% Canon, 17% Nikon, 11% Sony, … • 519k users with country information • 39% USA, 9% UK, 5% Canada, …, 27% unmatched • 11M directed edges with brand information • 1785 models, 96 brands, 168 countries

  8. Methodology: Pairwise Brand Congruence • Look at user pairs • X is in the list of contacts of Y (“friends”) • X and Y are random users (“baseline”) • X and Y are friends/random pairs with property Z • Percentage of congruent pairs • Congruent = same brand used • High congruence itself is not enough • Is the percentage for friends higher than for baseline

  9. Dependence on Friendship and Country Friendship matters ... ... more than country.

  10. Dependence on Closeness of Friendship “close” = similar interests = similar groups joined X 2 {G1,G2,G4}, Y 2 {G2,G3,G4,G6}, GJ(X,Y) = |{G2,G4}|/|{G1,G2,G3,G4,G5,G6}| = 2/6 Groups are irrelevant. “close” = mutual friends Mutuality is irrelevant. “close” = few friends (up to five) Friendship size matters. big difference no difference

  11. Dependence on Closeness of Friendship “close” = cliqued D {X,A,B,Y} {Y,B,C,D} Cliqueness matters. A X Y C B FJ(X,Y) = |{B,Y}|/|{A,B,C,D,X,Y}| = 2/6 big difference

  12. Dependence on Camera Type Point & Shoot (P&S) = cheap, used by “beginner” users Digital Single Lens Reflex (DSLR) = expensive, used by “expert” users no huge difference no huge difference Camera type matters. big difference big difference

  13. “Triggering” of Brand/Model Changes • Given a user changes her model 2007 -> 2008 • 54% high / 51% low cliqueness also change • 48% of random users change • Given a model change of user and friend • 38% high / 29% low cliqueness change to same brand • c.f. 33% congruent high cliqueness friends in 2008 • Given a model change of random users • 20% change to same brand • c.f. 19% congruent in 2008 • Country information only added 1-2% There seems to be some “triggering”.

  14. Possible Extensions • Use comments, view counts • weight friendship links, identify key users • More local analysis • city, age, particular brand, more fine-grained time • Other product networks • pdas/phones, cars, fashion, ...

  15. Gràcies! / Gracias!

  16. Additional Slides

  17. Cliqueness, Country, Frienship Size

More Related