language of politics on twitter - 03 analysis
TRANSCRIPT
Language of Politics on TwitterSummer School in AI
American University BeirutJune 16, 2015
Yelena Mejova@yelenammSocial Computing GroupQatar Computing Research Institute, HBKU
Roadmap
• lets talk politics (sampling)• political leaning– human classification– text-based classification– network-based classification
• look who’s talking (users)• predicting elections!
US politics
• Most research done so far• Clear left/right distinction• Popular political figures• High(ish) Twitter engagement REPUBLICAN
(right)DEMOCRAT
(left)
• Sampling Twitter for political speech– general keywords: #current– event keywords: #debate08, #tweetdebate– people: obama, romney, merkel– parties: democrat, republican, pirate– accounts: wefollow, twellow– news stories, known URL retweets
• Caveats– requires expert knowledge– known best after the event– selection bias (who do you want to ignore?)
topical sampling
bootstrapping
1. start with a few key words2. find tweets that have these words3. get more words out of these tweets
• seed sample with known political hashtags– #p2 – Progressives 2.0– #tcot – Top Conservatives on Twitter
• find hashtags which co-occurred with them, using Jaccard similarity
bootstrapping
tweets mentioning both
tweets mentioning either
bootstrapping
Predicting the political alignment of twitter users @vagabondjack Conover et al. @ SocialCom (2011)
got your #tag!
hashtag week party
aggregated user volume for (h,w)aggregated user volume for (*,w)• Given set of users with known leaning:
Political hashtag hijacking in the US Hadgu, Garimella, Weber @ WWW (2013)
[some figures from authors’ original slides]
Crimean conflict
Крымcomparing tweets by users withUkrainian or Russian as profile language
most distinguishing hashtags
Language Plurality in Twitter Political Speech Mejova, Boynton @ ICCSS (2015)
1. Crowdsourcing2. Text (text classification)3. Network (label propagation)
political leaning classification
crowdsourcing
• break the task into micro-tasks (N/Y question)• have many people answer for a bit of money• wisdom of crowds will give the right answer
Representing Text
• “Bag of words”, i.e. Vector Space Model
break the document into its constituent words and put them in a table
Representing Text
• Preprocessing– Clean-up• remove formatting, tables, HTML…
– Remove stopwords• the, of, to, a, in, and, that, for, is
– Stem words• get to a “stem” of a word• cats -> cat, running -> run, uncomfortable -> uncomfort?
Representing Text
• Vector Space Model:
those lazy cats sleep and sleep everywhere
D = (t1, wd1; t2, wd2; …, tv, wdv)
w: binary, count, TFIDF
lazy cat sleep everywhere …
1 1 2 1 …
Problems
• Synonymy– multiple words that have similar meanings
• Polysemy– words that have more than one meaning
EYE DROPS OFF SHELFPROSTITUTES APPEAL TO POPE
KIDS MAKE NUTRITIOUS SNACKSSTOLEN PAINTING FOUND BY TREE
LUNG CANCER IN WOMEN MUSHROOMSQUEEN MARY HAVING BOTTOM SCRAPEDDEALERS WILL HEAR CAR TALK AT NOONMINERS REFUSE TO WORK AFTER DEATH
MILK DRINKERS ARE TURNING TO POWDERDRUNK GETS NINE MONTHS IN VIOLIN CASE
GRANDMOTHER OF EIGHT MAKES HOLE IN ONEHOSPITALS ARE SUED BY 7 FOOT DOCTORS
LAWMEN FROM MEXICO BARBECUE GUESTSTWO SOVIET SHIPS COLLIDE, ONE DIES
ENRAGED COW INJURES FARMER WITH AXLACK OF BRAINS HINDERS RESEARCH
RED TAPE HOLDS UP NEW BRIDGESQUAD HELPS DOG BITE VICTIM
IRAQI HEAD SEEKS ARMSHERSHEY BARS PROTEST
• is it spam?• is it important?• is it happy?• is it true?• is it a flight ticket?
classifier
documentlabel
• is it written well?• is it about politics?• is it a bully?• is it fake?• is it a joke?
classifiers
naïve bayesdecision trees
support vector machineslogistic regression
perceptronneural networks
k-nearest neighbor
naïve bayes classifier
• We want to know probability of a class given an instance represented by a feature vector. By Bayes’ Theorem:
https://en.wikipedia.org/wiki/Naive_Bayes_classifier
constant no matter C
joint probability
naïve bayes classifier
• Expand the joint probability using the chain rule
• But to simplify, we use a naïve assumption of conditional independence for each feature
https://en.wikipedia.org/wiki/Naive_Bayes_classifier
naïve bayes classifier
• Finally, the conditional distribution over class C
scaling factor
probability of class C given a document with some features
prior of the class
frequency based probability of features in that class C
support vector machine
• Finds a hyperplane in high-dimensional space that maximizes the distance to the nearest training point of any class
https://en.wikipedia.org/wiki/Support_vector_machine
political leaning classification
Predicting the political alignment of twitter users @vagabondjack Conover,
Gonçalves, Ratkiewicz, Flammini, Menczer @
SocialCom (2011)
Is a user politically left or right?
actual classAB
predicted classA Bconfusion matrix
Classifier: Support Vector Machine
• Label propagation– Initialize cluster
membership arbitrarily– Iteratively update each
node’s label according to the majority of its neighbors
– Ties are broken randomly• Cluster assignment by
majority cluster label (using manually labeled data)
political leaning classification
retweet network
Twitter polarity classification with label propagation over lexical links and the follower graph
@speriosu Speriosu, Sudan, Upadhyay, Baldridge @ EMNLP (2011)
political leaning classificationkn
own
know
n
automaticallylabeled
news polarizationVisualizing media bias through Twitter
@JisunAn An, Cha, Gummadi, Crowcroft, Quercia @ AAAI (2012)
Jaccard similarity of their audience (co-subscribers)
distance between two media
overlap in common audience (followers on Twitter)
look who’s talkingVocal Minority versus Silent Majority:
Discovering the Opinions of the Long Tail @enimust Mustafaraj, Finn, Whitlock, Metaxas @ SocialCom (2011)
number of tweets per user
look who’s talking
Vocal Minority versus Silent Majority: Discovering the Opinions of the Long Tail
@enimust Mustafaraj, Finn, Whitlock, Metaxas @ SocialCom (2011)
GOP primary season on twitter: popular political sentiment in social media @yelenamm Mejova, Srinivasan, Boynton @ WSDM (2013)
look who’s talking
• Truthiness is a quality characterizing a "truth" that a person making an argument or assertion claims to know intuitively "from the gut" or because it "feels right" without regard to evidence, logic, intellectual examination, or facts.
Detecting and Tracking Political Abuse in Social Media Ratkiewicz, Conover, Meiss, Goncalves, Flammini, Menczer @ ICWSM (2011)
look who’s talking
Classifying memes (hashtags) for astroturf (fake grass roots movements)
Detecting and Tracking Political Abuse in Social Media Ratkiewicz, Conover, Meiss, Goncalves, Flammini, Menczer @ ICWSM (2011)
look who’s talking
most useful:network features
Truthy project by Indiana Universityhttp://truthy.indiana.edu/
look who’s talking
look who’s talking
#ampat @PeaceKaren_25 &@HopeMarie_25
gopleader.gov Chris Coons
#Truthy @senjohnmccain on.cnn.com/aVMu5y “Obama said…”
TRU
THY
LEG
ITIM
ATE
classifier
tweeton a topic
positive vs negative
Trained Classifiers Sentiment Lexiconscan “tune” for specific topic and data
but expensivecan use “out of the box”
but may not work for every topic
political discussions: debates
• Mean valence:– Obama: -2.09– McCain: -5.64
Characterizing Debate Performance via Aggregated Twitter Sentiment @ndiakopoulos Diakopoulos, Shamma
@ CHI (2010)
an emotional story
volume positive - negative
• 2009 German federal elections
electionsPredicting Elections with Twitter: What 140 Characters Reveal about Political Sentiment
Tumasjan, Sprenger, Sandner, Welpe @ AAAI (2010)
“The mere number of tweets reflects voter preferences and comes close to traditional election polls”
CONTROVERSY!
electionsWhy the Pirate Party won the German election of 2009 or the trouble with predictions: A
response to Tumasjan, Sprenger, Sander, & Welpe, "Predicting elections with twitter: What 140 characters reveal about political sentiment"
@ajungherr Jungherr, Jürgens, Schoen @ SSCR V30/N2 (2012)
“arbitrary choices”
If results of polls played a role in deciding upon the inclusion of particular parties, the TSSW method is dependent
on public opinion surveys
Choice of Parties Choice of Dates
prediction analysis […] between [13.9] and [27.9], the day of the election,
produces a MAE of of 2.13, significantly higher than the MAE for TSSW
• 2012 US Republican Primary Debates• Predicting polls swings around televised debates:
– 104 predictions overall
electionsGOP primary season on twitter: popular political sentiment in social media
@yelenamm Mejova, Srinivasan, Boynton @ WSDM (2013)
Both volume or sentiment classification are same than random
elections
single variable logistic regression models multi-variable logistic regression models
strong baselines!having followers (in your own party?)
focusing on centrist issues
graph structure and content significantly improve accuracy
The Party Is Over Here: Structure and Content in the 2010 Election Livne, Simmons, Adar, Adamic @ ICWSM (2011)
• Non-US elections:
– Irish: On using twitter to monitor political sentiment and predict election results, Bermingham, Smeaton (2011)• "Our approach however has demonstrated an error which is not competitive
with the traditional polling methods.”
– Dutch: Predicting the 2011 Dutch senate election results with twitter, Sang, Bos (2012)• Uses polls for demographic imbalances, yet performance still below
traditional polls
– Singapore: Tweets and votes: A study of the 2011 singapore general election, Skoric, Poor, Achananuparp, Lim, Jiang (2012)• Not as accurate as traditional polls, performance at local government levels
– many more coming out each day!
elections
• Data from social media are fundamentally different than data from natural phenomena– people change their behavior next time around– spammers & activists will try to take advantage
• From a testable theory on why and when it predicts (avoid self-deception!)
• (maybe) Learn from professional pollsters– tweet ≠ user– user ≠ eligible voter– eligible voter ≠ voter
How (Not) To Predict Elections @takis_metaxas Metaxas et al. @ SocialCom (2011)
elections
but what can we do?
help campaigners reach more peoplepredict people’s political leaning
help understand reasons for affiliationrecommend politicians, news, friends
detect sudden strong sentiment about a topicdetect polarization (users & news)
views of issues from around the world
• M. D. Conover, B. Gonçalves, J. Ratkiewicz, A. Flammini, and F. Menczer, “Predicting the political alignment of twitter users,” in Privacy, security, risk and trust (passat), 2011 IEEE Third International Conference on Social Computing (SocialCom), 2011, pp. 192–199.
• M. D. Conover, J. Ratkiewicz, M. Francisco, B. Goncalves, F. Menczer, and A. Flammini, “Political Polarization on Twitter,” International Conference on Weblogs and Social Media (ICWSM), 2011.
• M. Speriosu, N. Sudan, S. Upadhyay, and J. Baldridge, “Twitter polarity classification with label propagation over lexical links and the follower graph,” in Proceedings of the First workshop on Unsupervised Learning in NLP, 2011, pp. 53–63.
• I. Weber, V. R. K. Garimella, and A. Teka, “Political hashtag trends,” in in Advances in Information Retrieval, Springer, 2013, pp. 857–860.
• A. T. Hadgu, K. Garimella, and I. Weber, “Political hashtag hijacking in the US,” in Proceedings of the 22nd international conference on World Wide Web companion, 2013, pp. 55–56.
• M. Pennacchiotti and A.-M. Popescu, “Democrats, republicans and starbucks afficionados: user classification in twitter,” in Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining, 2011, pp. 430–438.
• N. A. Diakopoulos and D. A. Shamma, “Characterizing Debate Performance via Aggregated Twitter Sentiment,” Conference on Human Factors in Computing Systems (CHI), 2010.
• L. Chen, W. Wang, and A. P. Sheth, “Are twitter users equal in predicting elections? a study of user groups in predicting 2012 US republican presidential primaries,” in Social Informatics, Springer, 2012, pp. 379–392.
• J. An, M. Cha, K. P. Gummadi, J. Crowcroft, and D. Quercia, “Visualizing media bias through Twitter,” Association for the Advancement of Artificial Intelligence (AAAI), Technical WS-12-11, 2012.
• E. Mustafaraj, S. Finn, C. Whitlock, and P. T. Metaxas, “Vocal Minority versus Silent Majority: Discovering the Opinions of the Long Tail,” in International Conference on Social Computing, 2011, pp. 103–110.
• J. Ratkiewicz, M. D. Conover, M. Meiss, B. Goncalves, A. Flammini, and F. M. Menczer, “Detecting and Tracking Political Abuse in Social Media,” International Conference on Weblogs and Social Media (ICWSM), 2011.
• A. Livne, M. Simmons, E. Adar, and L. Adamic, “The Party Is Over Here: Structure and Content in the 2010 Election,” International Conference on Weblogs and Social Media (ICWSM), 2011.
• A. Tumasjan, T. O. Sprenger, P. G. Sandner, and I. M. Welpe, “Predicting Elections with Twitter: What 140 Characters Reveal about Political Sentiment,” Association for the Advancement of Artificial Intelligence Conference (AAAI), 2010.
• P. Metaxas, E. Mustafaraj, and D. Gayo-Avello, “How (Not) To Predict Elections,” International Conference on Social Computing, 2011.
• A. Jungherr, P. Jürgens, and H. Schoen, “Why the pirate party won the german election of 2009 or the trouble with predictions: A response to Tumasjan, a., Sprenger, to, Sander, pg, & Welpe, im ‘predicting elections with twitter: What 140 characters reveal about political sentiment’,” Social Science Computer Review, vol. 30, no. 2, pp. 229–234, 2012.
• I. Weber, V. R. K. Garimella, and A. Batayneh, “Secular vs. Islamist polarization in Egypt on Twitter.” ASONAM, 2013.
Surveys• D. Gayo-Avello, “‘ I Wanted to Predict Elections with Twitter and all I got was this
Lousy Paper’--A Balanced Survey on Election Prediction using Twitter Data,” arXiv preprint arXiv:1204.6441, 2012.
• D. Gayo-Avello, “A meta-analysis of state-of-the-art electoral prediction from Twitter data,” Social Science Computer Review, 2013.