• Two metaphors for the web:
    • The library (knowledge, pages, hyperlinks, associations)
    • The crowd (real-time awareness, memes, contagion)
  • Strong ties and weak ties in social net are distinct w/r/t what flows across edges
    • Strong ties: trusted interactions, social capital
    • Weak ties: access to new information
  • What are online social networks accomplishing for their users?
    • A social transport layer for information
    • Helping maintain ties that are too weak for you to maintain on your own
  • Issue: networks have low diameter but trees are deep
    • Open problem: a reasonable model of tree depth with probable guarantees
  • How do node properties vary as we move down a cascade tree?
    • What is prob that two random trees agree on attribute?
    • Relationship to ancestor reconstruction problems
  • How does alignment of attributes relate to adoption rate?
    • Study approx 2300 most popular apps on Facebook
    • Look @ how homophily affects adoption
    • 75% of apps on Facebook are homophilous - more likely to accept invitation from similar friend
    • Question: should strategy for marketing app depend on its homophily parameter?
  • Characterizing types of social ties:
    • Given a person's neighborhood, can we identify their most significant social ties
    • Core componenet in prioritizing content
    • Rank neighbors by embeddedness?
    • Embeddedness of e: num of mutual friends of e's endpoints
    • In practice, embeddedness finds many nodes from the largest cluster
    • Challenge problem: Can you find the relationship partner of a Facebook user given their network structure alone?
    • Looked at 1.3 million Facebook profiles that are in relationship
    • Embeddedness and Dispersion vs Num Photos where they've both been tagged and number of mutual profile views
    • Generally, dispersion does better than other features @ prediction of relationship
    • Combining all through ML does better
    • Signal stronger w/ age of relationship