Ego networks and centrality

Christophe Prieur,
Télécom ParisTech, dept. SES, I3

GDR Réseaux SHS

Rouen

oct 2016

Interaction / relationship

By phone

Contacts ranked by frequency

Saramaki, [Dunbar] et al, 2014, on "social signatures"

On Facebook

  • Marlow, 2009
  • Burke, Kraut, Marlow, 2011
  • Jones et al, 2013
  • Burke & Kraut, 2013, 2014, 2016
  • From large-scale Facebook data
  • most online interaction with face-to-face contacts
  • comments > likes > private messages > photo tags,
    best predictors for tie strength

the Algopol Facebook Survey

  • in collaboration with CSA poll agency
  • a sample of 880 representative internet users
  • 15 000 respondents
  • large-scale personal network analysis
  • ethnographic interviews

Facebook ties

talking religion or politics?

A Facebook app

app.algopol.fr

data-based interviews

Proximity / centrality

  • likes + com / proximity + frequency (Jones, Burke, Kraut, et al)
  • degree, without betweenness: central in a dense cluster, not necessarily close
  • betweenness / proximity + interaction, but not necessarily friend

Significant alter

Dispersion

Backstrom & Kleinberg, 2014
about couples in Facebook ego-networks

  • Given ego and alter,
  • take ego's network without ego and alter,
  • dispersion (u,v)
    = sum of distances between all pairs of nodes
  • distance (u,v)
    = 1 if u and v (friends of) friends of each other,
    0 otherwise
  • predicts if ego is
    "in a relationship"
  • if yes, guesses the lover with 50%

    (only 25% when using shared friends)

  • when bad dispersion value, twice as much likely to break within 2 months
  • 525 égos ont qualifié leur compagnon / conjoint
  • pour 264 (soit 50%), il est l'alter le plus central
    • parmi eux, 62 sont mariés
    • parmi les autres 50%, seuls 17 sont mariés

How central is the significant other?

  • rank betweenness of alters
  • ratio first / second values of betweenness
en médiane, pour tous les égos déclarés mariés sur Facebook, l'alter le plus central est 2.5 fois plus central que le deuxième

Clusters

Two regimes of conversation

Two regimes of conversation

Of borders,
bridges and doors...

Enumerating graphlets

Representativity

  • Overall representativity:
    17,195,648 / 670,586,458
    = 0.0256
  • occurrences here: 22,303
  • total number of graphlets here: 2,354,731
  • representativity:
    (22,303 / 2,354,731) / 0.0256
    = 0.37