PyData 2013 Tutorial:

Social Network Analysis

~

Katherine Chuang

@katychuang

1. Introduction to SNA Theory

SNA is a powerful research technique.

1. Tracking health epidemics (i.e. spread of infections)
2. Measuring an online community (i.e. conversation patterns)
3. Graphically display data

SNA is essentially graph theory. A network is made up of nodes and edges.

Edges can have direction.

Edges can be weighted

Nodes can be of a type (social roles)

Describing individual nodes

• Centrality - an individual node's number of links
• Degree – number of links from and to the node.
• Betweenness – The extent to which an individual node is 'between' other nodes
• Closeness – The degree to which an individual is near all other individuals in a network (directly or indirectly).

Describing subgroups

• Cliques - a group where every node is directly connected to each other
• Clustering Coefficient - the average density of a subgroup
• Blockmodels - a cluster of users based on type of relationship, i.e. novices
• Eigenvector Centrality - highlights the importance of users in a network

Yellow = experts answering questions; Red = novices

http://lithosphere.lithium.com/t5/science-of-social-blog/
Social-Graphs-The-Art-and-the-Insights/ba-p/5713

Lots of dialogue back and forth between users

http://lithosphere.lithium.com/t5/science-of-social-blog/
Social-Graphs-The-Art-and-the-Insights/ba-p/5713

Q&A + Discussion

http://lithosphere.lithium.com/t5/science-of-social-blog/
Social-Graphs-The-Art-and-the-Insights/ba-p/5713

Experts are scattered in community

http://lithosphere.lithium.com/t5/science-of-social-blog/
Social-Graphs-The-Art-and-the-Insights/ba-p/5713

Social Interaction Design

UI design for self-presentation

and interacting with others

It's your social history...

Design patterns

• Discussion threads, reviews, classified listing, news, data sharing
• Responding via likes, votes, free text
• Each unit of analysis has a sender and receiver (2 nodes, 1 edge)

Recap: Map online conversations

please ensure you have Python v2.7 and Networkx v1.7

$ pip freeze

# alternatively, if you do not have these installed,
# you can create an account on pythonanywhere.com

2. Introduction to NetworkX

NetworkX v1.7

Workflow

1. Read in raw data
2. NetworkX magic
3. Export image

NetworkX basics

import networkx                         # import library

g = networkx.Graph()                    # create undirected graph object

g.add_edge(1,2)                         # add data
g.add_edge(3,1)
g.add_node(4)

print g.number_of_nodes()               # 4
print g.nodes()                         # [1, 2, 3, 4]

print g.number_of_edges()               # 2
print g.edges()                         # [(1, 2), (1, 3)]

NetworkX Directed Graph

g = networkx.DiGraph()  # directed graph

g.add_edges_from([("A","B"), ("C","A")])

print g.in_degree(with_labels=True) # {'A': 1, 'C': 0, 'B': 1}

print g.out_degree(with_labels=True) # {'A': 1, 'C': 1, 'B': 0}

print g.neighbors("A")  # ['B']
print g.neighbors("B") #['A']

NetworkX modules

print g.degree()        # {1: 2, 2: 1, 3: 1, 4: 0}

print networkx.betweenness_centrality(g)
                        # {1: 0.3333333333333333, 2: 0.0, 3: 0.0, 4: 0.0}

print networkx.degree_centrality(g)
                        # {1: 0.6666666666666666, 2: 0.3333333333333333,
                        # 3: 0.3333333333333333, 4: 0.0}

Time for magic tricks.

# Verify your python installation
$ python --version

# Make sure networkx can be used.
$ python -c "import networkx;"

# Open up your text editor

3. Let's code!

Open up a text editor

PythonAnywhere.com has one built in.

Exercise..

• Importing data
• Plot data points
• Export images

Download: http://bit.ly/15WDFW9

# Script:
https://github.com/katychuang/PyData2013-SNATutorial/blob/master/virus.py

# Documentation:
http://networkx.github.com/documentation/latest/contents.html

# MatPlotLib:
http://matplotlib.org/users/gridspec.html

"Social interaction design is the application of levers to steer and guide emerging social practices" -- Adrian Chan, Gravity7

Questions?

• Email: [email protected]
• Twitter: @katychuang
• Code: github.com/katychuang