Political discussions on social network platforms represent an increasingly relevant source of political information, an opportunity for the exchange of opinions and a popular source of quotes for media outlets. We analyzed political communication on Twitter during the run-up to the German general election of 2009 by extracting a directed network of user interactions based on the exchange of political information and opinions. In consonance with expectations from previous research, the resulting network exhibits small-world properties, lending itself to fast and efficient information diffusion. We go on to demonstrate that precisely the highly connected nodes, characteristic for small-world networks, are in a position to exert strong, selective influence on the information passed within the network. We use a metric based on entropy to identify these New Gatekeepers and their impact on the information flow. Finally, we perform an analysis of their input and output of political messages. It is shown that both the New Gatekeepers and ordinary users tend to filter political content on Twitter based on their personal preferences. Thus, we show that political communication on Twitter is at the same time highly dependent on a small number of users, critically positioned in the structure of the network, as well as biased by their own political perspectives.

1. Small Worlds with a
Difference
New Gatekeepers and the Filtering of Political
Information on Twitter
full paper at the websci repository
Pascal Jürgens Andreas Jungherr Harald Schoen
University of Bamberg University of Bamberg
University of Mainz
andreas.jungherr harald.schoen
pascal.juergens@uni-mainz.de
@uni-bamberg.de @uni-bamberg.de
1

2. Political Tweets
(German general election 2009)
“Ihr werdet euch noch wünschen wir wären
Politikverdrossen [sic!].” – Max Winde (@343max)
rough translation:
“Soon you’ll be wishing we were through with
politics”
2

3. A Political Question
Assuming that …
the web is a considerable source for political
information
each user only perceives a selection of this
information
facts and opinions have the potential to influence
recipients and journalists
Q: what forces influence the visibility of information?
3

4. Old Answer
Journalists as gatekeepers
vs
Recipients as selective readers
(“Uses & Gratifications”)
4

5. New Environment
Networked Information
Relevant – social networking sites see
immense growth in adoption
New paradigms – real-time, recommended
many-to-many communication
Perceived information varies, selection is
distributed
Old answers neglect network effects in the process
5

6. What We Know
Social networks tend to exhibit small world
properties (so do networks of political communication on twitter)
Information flows fast and networks are
relatively resistant to (random) disruptions
Information should be able to move unencumbered
6

7. The Problems
Contagion model of information spread
Links can be substituted
But: Almost everyone has almost no friends (power
law), key players have dominant role
Information relay via key players
Non-Random removal!
Problem moves from contagion to disruption
(Borgatti’s KPP-Neg problem)
7

8. Impact of Key Player
Removal
Comput Math Organiz Theor (2006) 12: 21–34 25
Fig. 3 Network in which removing the two most central nodes (“i” and “ j”) is not as disruptive as removing
a different pair of nodes (“i” and “ j”) Borgatti (2010) p.39
8

9. Impact of Key Player
Removal
Comput Math Organiz Theor (2006) 12: 21–34 25
Fig. 3 Network in which removing the two most central nodes (“i” and “ j”) is not as disruptive as removing
a different pair of nodes (“i” and “ j”) Borgatti (2010) p.39
8

10. Impact of Key Player
Removal
Comput Math Organiz Theor (2006) 12: 21–34 25
Network of 8,609 german Twitter users who
used political hashtags during the general
election campaign 2009 (June 18 to September 30, 2009)
Edges are directed messages (RT and @) which
contain political hashtags
Fig. 3 Network in which removing the two most central nodes (“i” and “ j”) is not as disruptive as removing
a different pair of nodes (“i” and “ j”) Borgatti (2010) p.39
8

11. Finding Key Players
Need general measure for a node’s impact on
network’s capability to transfer information
» Ortiz-Arroyo (2010):
Centrality Entropy metric (Hce)
9

12. Centrality Entropy
Calculates the entropy of a network = capacity
for information transfer (~ease of transfer)
Iteratively remove nodes and re-calculate to
determine largest impact = key players*
(KPP-NEG)
Complexity is O(n3)
*note: no optimal solution for KPP-Neg
10

13. Results
Network centrality entropy: 12.1596
The most influential node reduces entropy by ~.1
when removed (N= 8,609)
Highest entropy impact of a node in Borgatti’s
example networks: ~.1 (N=19)
Entropy impact declines rapidly (power law?)
» A small number of users can have a
disproportionally large disruptive impact
11

14. Political Tweets II
During the election campaign, users explicitly
coded party affiliations:
#party+ = positive (e.g. cdu+, spd+)
#party– = negative (e.g. cdu–, spd–)
This allows for the creation of party affiliation
profiles
12

15. User Bias
We calculated the distribution of party
evaluations for outgoing and incoming*
directed messages
A simple χ2 test was used to test for difference
of those distributions
*messages from network neighbors
13

16. Results
A majority of users with testable volume of
messages displayed a significant (p < 0.001) bias
All testable gatekeepers (top100) displayed bias
sig. bias no sig. bias
negative party
2,866 201
evaluations
positive party
3,835 34
evaluations
Users with significantly differing distributions of
incoming and outgoing party evaluations. N = 8,609
14

17. Example User
ased the network’s
ustrate the destructive Table 2. Political bias of the Twitter account @volker_beck by
mber of strongly and the German politician Volker Beck (Bündnis 90/Die Grünen)
om 5881 to 6574 and
time, the average path Outgoing2 Incoming3
CDU (conservatives) 0 511
urn to the question of
CSU (conservatives) 0 58
ad in the network. In
epers were shown to SPD (social democrats) 0 876
ty in their stories in
mical selection criteria FDP (liberals) 0 1,269
gard to the political Grüne (green party) 19 630
and New Gatekeepers
political position more Linke (socialists) 0 267
ose messages that the
published or chose to Piraten (pirate party) 10 4,944
ntly skewed from the
y receive? In short, do
around a user, or are Of the users in the conversation network, roughly 3,000 had
ased on choices by the enough in- and outgoing Twitter messages to statistically test for a
allows for a precise bias. For negative mentions of parties, 2,866 users had a
significant bias (p < .001) while a mere 201 users did not exhibit a
significant bias. It should be noted that a bias could be identified
15
rs coded their Twitter

18. Takeaways
The network of political conversations during
the general election campaign 2009 was
dominated by a small number of key users
which functioned as critical relays within the
network
16

19. Takeaways
In their tweets, these key users did not mirror
their network environment but instead
exhibited an individual political bias
17

20. Takeaways
The visibility of political information on twitter
is critically dependent on the network
structure
18

21. Thank you for your (early morning)
attention!
Key References
Bakshy, E, Hofman, J. M., Mason, W. A., and Watts, D. J. 2011. Everyone’s an Influencer: Quantifying Influence on Twitter. Proceedings of
the Fourth International Conference on Web Search and Data Mining (Hong Kong, China, February 09 – 12, 2011). WSDM 2011.
Borgatti, S. P.,2005. Centrality and network flow. Social Networks 27, 55-71. DOI - http://dx.doi.org/10.1016/j.socnet.2004.11.008.
Borgatti, S. P. 2006. Identifying sets of key players in a social network. Computational & Mathematical Organization Theory 12 (1),
21-34. DOI- http://dx.doi.org/10.1007/s10588-006-7084-x.
Cha, M., Haddadi, H., Benevenuto, F., and Gummadi, K. P. 2010. Measuring User Influence in Twitter: The Million Follower Fallacy. In
Proceedings of the Fourth International AAAI Conference on Weblogs and Social Media (Washington, DC, May 23 – 26, 2010).
ICWSM. AAAI Press, Menlo Park, CA, 10-17.
Jungherr, A., Jürgens, P., and Schoen, H., 2012. Why the Pirate Party Won the German Election of 2009 or The Trouble With
Predictions: A Response to Tumasjan, A., Sprenger, T. O., Sander, P. G., & Welpe, I. M. ‘Predicting Elections With Twitter: What 140
Characters Reveal About Political Sentiment’. Social Science Computer Review. Forthcoming. DOI- http://dx.doi.org/
10.1177/0894439311404119.
Jürgens, P., and Jungherr, A. 2011. Wahlkampf vom Sofa aus: Twitter im Bundestagswahlkampf 2009. In: Schweitzer, E. J., and Albrecht, S.
(eds.). Das Internet im Wahlkampf: Analysen zur Bundestagswahl 2009.VS-Verlag, Wiesbaden, 201-225. DOI- http://dx.doi.org/
10.1007/978-3-531-92853- 1_8.
Ortiz-Arroyo, D. 2010. Discovering Sets of Key Players in Social Networks. In: Abraham, A., Hassanien, A.-E., and Snásel ,V. (eds.).
Computational Social Network Analysis. Springer Verlag, Dordrecht et al., 27-46. DOI- http://dx.doi.org/
10.1007/978-1-84882-229-0_2.
Shannon, C.E. 1948. A mathematical theory of communication. The Bell System Technical Journal 27, 379– 423, 623–656.
Watts, D. J., and Strogatz, S. H. 1998. Collective dynamics of ‘small-world‘ networks. Nature 393, 440-442. DOI- http://dx.doi.org/
10.1038/30918.
19