Injustice - Developers Among Us (SciFiDevCon 2024)
Visualizing Co-Retweeting Behavior for Real-Time Debate Recommendations
1. S A M A N T H A F I N N A N D E N I M U S T A F A R A J
W E L L E S L E Y C O L L E G E
P R E S E N T E D A T M S M 2 0 1 3
C O - L O C A T E D W I T H A C M H Y P E R T E X T 2 0 1 3
P A R I S , F R A N C E
Visualizing Co-Retweeting
Behavior for Recommending
Relevant Real-Time Content
2. “Thomas Jefferson used newspapers to win the
presidency, F.D.R. used radio to change the way he
governed, J.F.K. was the first president to understand
television, and Howard Dean saw the value of the
Web for raising money…
But Senator Barack Obama understood that you
could use the Web to lower the cost of building a
political brand, create a sense of connection and
engagement, and dispense with the command and
control method of governing to allow people to self-
organize to do the work.”
- New York Times Article, How Obama Tapped Into Social Networks’ Power
How to become the US President
3. The US Presidential Debates
Debates important events in the election
Second only to Super Bowl in TV watch
Since the first televised debates in 1960, many
defining moments people remember from the
debates
Al Gore rolling his eyes and sighing
Reagan “There you go again”
Bush Sr. looking at his watch
6. Source: Dispatch From the Denver Debate (blog.twitter.com)
Watching on Twitter: First Debate
7. Watching on Twitter: Second Debate
Source: Twitter at the Town Hall Debate (blog.twitter.com)
8. Watching on Twitter: Third Debate
Source: The Final 2012 Presidential Debate (blog.twitter.com)
9. Follow the Debate Through Hashtags (Or try to)
Interactive Chart
There are too many hashtags being created, and rising and falling in popularity
during the debate to follow them all.
10. User Number Followers Tweets during the
3rd debate
ladygaga 30,572,024 2
BarackObama 21,206,234 9
YouTube 18,484,946 1
twitter 13,993,216 1
JimCarrey 9,256,715 1
cnnbrk 9,051,349 4
nytimes 6,350,936 8
CNN 6,287,257 5
PerezHilton 5,742,938 8
MTV 5,725,897 2
Or by Following Famous Twitter Accounts
Popular, celebrity Twitter accounts are
NOT the ones generating content during the debates.
12. Single person has a limited view
It would be impossible for a single person to discover
and consume all the content about the debates
(especially while trying to watch them)
Solution?
Human computation
13. Human Computation
Using humans as computers to improve intelligent
algorithms
Retweeting as recommendation
Already existing Twitter construct
Utilizing it for a new purpose
14. uA uB uC uD
u1 u2 u3 u4 u5
. . . . .
. . . . .
. . . . .
(tweeters)
(tweets)
(retweeters)
Going from a tweeting model…
15. To the Retweet Matrix…
uA uB uC uD
u1 2 1 0 0
u2 0 1 0 0
u3 1 0 1 0
u4 0 1 1 1
u5 0 0 1 2
Tweeting Users (items)
RetweetingUsers(users)
16. uA uB uC uD
uA 2 1 1 0
uB 3 1 1
uC 3 2
uD 2
To the Co-Retweet Matrix
Tweeting Users
TweetingUsers
17. Co-Retweet Visualization
Nodes represent top
retweeted accounts
uA – uD
Edges mean that the
connected nodes have
been co-retweeted
Weighted by how many
users have co-retweeted
the two nodes
Created using Gephi
18. Step 1: Layout
Force Atlas
Algorithm developed by
Gephi
Force Directed layout
Attraction between
connected nodes
Repulsion between
unconnected nodes
Nodes with stronger
connections (more edges,
heavier weight) are
attracted to each other
Source: ForceAtlas2, A Graph Layout Algorithm for
Handy Network Visualization
19. Step 2: Community
Modularity Algorithm to
assign groups
Detects communities
within a network
Each community has a
different color
Source Fast unfolding of communities
in large networks
20. Step 3: Node Rank
Eigenvector centrality
Measures the influence of
a node in the network
based off the connections
with that node
Similar to Google’s
PageRank algorithm
Nodes are made larger
and darker based on
higher centrality values
23. Why a Co-Retweeting Model?
Co-retweeting focuses only on the content creators
Reveals the perceived
relationships between
accounts
Unbiased media accounts
24. Co-Retweet Network vs. Retweet Network
Retweet network
contains three
types of accounts
Users who generate
original content
Users who
aggregate content
by retweeting other
sources
Users who do both
Authorities vs.
Hubs
Retweeted
Only
48.0%
Wrote
Original
Tweets
Only
41.2%
Both Wrote
and
Retweeted
10.8%
25. Authorities Hubs
Accounts writing
tweets
Creating original
content
Accounts retweeting
content
Aggregating data from
lots of sources
Sifting through and
picking out content
they find worth
recommending
Comparing Content Creators to Retweeters
26. Applications of Recommender System
Given the set of users a person follows or has retweeted
Recommend additional users to follow during the
debates
Useful for people who are not savvy with Twitter
Don’t know the users to follow
Or don’t generally focus on political content, only interested during
election season
Focus on users who are creating interesting content
during the debates
Might not be the users with millions of followers
Or they don’t regularly tweet about politics
Have a record of the most interesting content of the
debate, to browse based on your own interests
27. Are these accounts still tweeting about politics?
Politically
Active
34.6%
Nonpolitical
65.4%
Of the top 1,500 retweeted accounts for the 16th and the 22nd, how many are
still generating relevant political content?
Source: Finn, S., and Mustafaraj, E. 2012. Learning to Discover Political Activism in the Twitterverse.
In KI-Künstliche Intelligenz 27 (1), 17-24
28. Conclusion
Massive amount of data on Twitter during the
debates
Human computation to create recommender system
Co-retweeting connections reveal perceived
relationships between accounts
Recommender system allows for easier and more
meaningful consumption of data in real time