Social Networks & Data

Got a chance to attend amazing talk by Prasanta Bhattacharaya (

at PyDataSG

Started with your favorite Game of Thrones network (I swear i still havent seen a single episode.)

Every betrayal ever…

Building Social network for Game of Thrones.

History Of Social Networks

  • Euler was probably the first one to give an algo for a social network. He came out with a theorem called Konigsberg 7 bridges problem in 1736.
  • The first data came after almost a century from Pierre Huber in 1802, with Social structure of Bumblebees.
  • MRT is an offline network. Recently Channel News asia published infographics for MRT delay vs mean distance travelled. Green line is doing worse.

How nature compares with human engineers.

Basic Network properties

  • Actors & relations <nodes /edges>
  • Relations <directed vs undirected ; binary vs valued(weighted)>

Storing network Data

  • Adjacency Matrices
  • Adjacency List
  • Edge List <most common form of network data>

Node Level Attributes

Degree Centrality

  1. Closeness Centrality : measure of distance of node from all other nodes
  2. Betweenness Centrality : Measure of all the shortest distance from a node
  3. Eigenvector Centrality : Measure of the influence of a node in a network.

(Google Pagerank uses eigenVector )

Research Challenges

Two broad area

  1. struct & evolution
  2. Process
  • why ppl make friend
  • what information spread
  • how disease spread

3. But Structure & processes are genereally cofounded

Favorite research area — “peer influence”

  • for a person i, intervention on Xi leads to outcome of Yi
  • But what about intervention on Yj where path (i,j) exists

Peer influence :

  1. If your friend buys iPhone, how soon will you buy one

2. Spread of disorders

3. Voting behavior

HOMOPHILY or Peer Influence

Hypothetical story from Solizi & Thomas (2011)

  • Two friends = Joey & Ian if Joey friend jumps of bridge , would Ian jump off too
  • Why would you jump off bridge (6 explanations)

Social Advertising

Showing social cue gets more clicks ?

Research paper from Facebook

Like Rate increases with D (number of peers shown), this is targeted Ad

and shows Homophily.

Usually social networks do A/B testing on social network does either at New Zealand or does Graph cluster randomization