Download - 2011 Spanish General Election
2011 Spanish General Elections
2012, July 6 M. Luz Congosto / Pablo Aragón 1
SUMMARY
Twitter on Election campaign
Status of art on Electoral prediction
Case study: 2011 Spanish General Election
Conclusions / Findings2012, July 6 M. Luz Congosto / Pablo Aragón 2
Twitter on Election campaign
2012, July 6 M. Luz Congosto / Pablo Aragón 3
Comunication Opinion Sensor
Data Mining Prediction
Status of art on Electoral prediction
(Tumasjan, 2010) 2009 German Elections. Method count mentions 1,65% MAE (Mean Absolute Error)
(Jungherr, 2011) 2009 German Elections (Conover D. , 2010) 2010 US Elections (Gayo-Avello D. , 2011) 2008 US elections (Tjong, 2012) 2011 Dutch Elections (Skoric, 2012) 2011 Singapur 2011 (Bermingham et al., 2011) 2011 IIreland Elections (Panagiotis, 2011) 2010 US Elections
2012, July 6 M. Luz Congosto / Pablo Aragón 4
Case study: 2011 Spanish General Election
Methodology
Twitter as a communication channel
Twitter as an opinion sensor
Twitter as a connection net
Twitter as a source of prediction 2012, July 6 M. Luz Congosto / Pablo Aragón 5
Case study: 2011 Spanish General Election
Methodology– Dataset 1: Monitored tweets with mentions of national parties
from 08/10/11 to 22/11/11 using Twitter streaming API from Carlos III University getting 2,973,110 tweets from 441,795 unique users
– Dataset 2: Stored tweets with mentions of political parties represented in Parliament from 9-10-2011 to 24-11-2011 using a routing process for downloading some users' timeline and Twitter streaming API from Fundació Barcelona Media getting 2,279,250 tweets from 442,014 unique users
2012, July 6 M. Luz Congosto / Pablo Aragón 6
Twitter as a communication channelCandidate vs. Party
2012, July 6 M. Luz Congosto / Pablo Aragón 7
Rajoy
RubalcabaCayoLara
PSOE PP
UpyD Equo
Treemap of followers before the campaing (Dataset-1)
2012, July 6 M. Luz Congosto / Pablo Aragón 8
Accumulated of tweets publish on campaign (Dataset-1)
Activity of campaign accounts on Twitter
Twitter as a communication channel
2012, July 6 M. Luz Congosto / Pablo Aragón 9
Accumulated of new followers on campaign(Dataset-1)
Getting new followers
Twitter as a communication channel
2012, July 6 M. Luz Congosto / Pablo Aragón 10
Timeline (Dataset-1)
Correlation new followers / unique mentions
Twitter as communication channel
Correlation by day (Dataset-1)
2012, July 6 M. Luz Congosto / Pablo Aragón 11
Tweets and users by day on campaign (Dataset-1)
Citizen Participation
Twitter as an opinion sensor
2012, July 6 M. Luz Congosto / Pablo Aragón 12
Valence by day of campaign (Dataset-2)
Emotionality (valence)
Twitter as an opinion sensor
2012, July 6 M. Luz Congosto / Pablo Aragón 13
Dominance by day on campaign (Dataset-2)
Emotionality (dominance)
Twitter as an opinion sensor
2012, July 6 M. Luz Congosto / Pablo Aragón 14
Treemap of mentions of Web sites on campaign (Dataset-1)Inteactive image: http://barriblog.com/taller/javascript/protovis/sites_20N.html
Spread links
Twitter as an opinion sensor
Política.El País
El País
Público
El mundo
EuropaPress
ABC
2012, July 6 M. Luz Congosto / Pablo Aragón 15
Mapa of RTs between politicians on campaign (Dataset-2)
User Communities
Twitter as a connection net
2012, July 6 M. Luz Congosto / Pablo Aragón 16
Mentions count on campaign (Dataset-1)
Mentions vs. Results
Twitter as a source of prediction
Total mentions (name + @user + #hashtag) MAE=1,66%
2012, July 6 M. Luz Congosto / Pablo Aragón 17
Users with more than three polarity RTs or #hashtags on campaign (Dataset-1)
Political Polarity vs. Results
Twitter as a source of prediction
Total users MAE: 5,00%
2012, July 6 M. Luz Congosto / Pablo Aragón 18
Political Polarity vs. Results
Twitter as a source of prediction
Men(61,38%) MAE: 6,49% Women (38,62%) MAE: 3,88%
Users with more than three polarity RTs or #hashtags on campaign (Dataset-1)
Conclusions / Findings
With measurements based on the mentions count we have got a good result, however:
The results depend on many factors such as the social-cultural environment in the elections, the period of the sample, campaign events, the collection of data on Twitter, the parties analyzed and calculation methodThe validation of this method of forecasting requires systemization of steps and checking of other elections
2012, July 6 M. Luz Congosto / Pablo Aragón 19
Conclusions / Findings
With the measurements based on Political polarity we have got worse results, we’ll have to bear this in mind to improve the algorithms: 1.Demographics: Twitter users are young and highly educated.
2.Hidden opinion: Not all users show their political opinions
3.Over opinion: Some parties supporters are very actives
4.Entity vs. People: It’s difficult to distinguish an entity from a person on Twitter
5.Anonymous vs. “Real Identity”: users with a real identity are more likely to have a hidden opinion
6.Men vs. Women: There is a gender difference. Men are likely to hide their opinion or to over opinion than women2012, July 6 M. Luz Congosto / Pablo Aragón 20