Literature Review - A Social Network Analysis Approach

Clement Lee

2018-08-24 (Fri)

Background

  • Someone asked “how do you do literature review?”

  • 130+ papers on social network analysis to read

  • Want to cluster papers according to their themes/methods

Network of papers

If A cites B, A is “connected” to B (one-way)

Two types of networks

  1. Co-authorship/collaboration network
    • The nodes are the authors
    • A & B are connected to each other if they have co-authored a paper
  2. Citation network
    • The nodes are the papers
    • Not really a social network

Community detection

Paper A B C D \(\cdots\)
Group 1 \(\surd\) \(\cdots\)
Group 2 \(\surd\) \(\surd\) \(\cdots\)
Group 3 \(\surd\) \(\cdots\)

  • Clustering the papers so that tightly connected ones belong to the same group

  • But each paper can only belong to one group at a time

  • What about interdisciplinary ones? Or reviews?

Mixed membership

  • Is this possible instead? Yes
Paper A B C D \(\cdots\)
Group 1 0.4 0.1 0.0 0.0 \(\cdots\)
Group 2 0.3 0.5 0.0 0.8 \(\cdots\)
Group 3 0.3 0.4 1.0 0.2 \(\cdots\)

Topic modelling

  • Analysing e.g. frequency of domain-specific terms

  • A paper belongs to multiple groups with weights

  • Assigning keywords in a quantitative/objective way

Community detection + mixed membership

  • Analysing the papers’ content citations to cluster them

  • A paper can belong to multiple groups

  • Tightly connected ones belong to have high weights in the same group

Results

Back to the network

Colouring by the primary group

Visualising group memberships

Summary

What it cannot do

  • Help you read the individual papers

What it can do

  • Cluster the papers in an objective way
  • Give you the big(ger) picture

Co-authorship network of … CHI papers

Publication count & OpenLab membership

Clustering & colouring

Next

  • Cluster with more groups & compare with known structure
  • Citation network of CHI papers
  • Co-authorship network of all OpenLab papers