Productgeruchten - PDF document

TINPOT ( Taal , Identiteit , Netwerken en Productgeruchten Op Twitter) From TweetGenie to the Apocalypse Theo Meder Dong Nguyen Rilana Gravel Corpus collection  User selection  Dutch Twitter users were sampled by searching on the query ‘ het ’ . An additional set was collected by randomly sampling from the followers/followees of the initial set of users.  Sampled in August 2012.  Annotation  Two annotators were employed to annotate properties such as gender, age, life stages, geographic location, type of avatar etc.  3185 users annotated. Language use on Twitter: Previous Research Gender - and age - based differences Gender and age prediction  A system that predicts the gender, exact age, age categories and life stages using logistic and linear regression (exact age) based on 200 tweets of a user  Automatic versus manual prediction.  Gender: Comparable performance  Age: Automatic system performed better than humans, while being much faster. For example, humans were on average 4.8 years of the exact age, the automatic system 4.1 years.  Harder to estimate the age for older persons. Red = women, Blue = men 10 20 30 40 50 60 0 . 0 0 0 . 0 4 0 . 0 8 Ag e P r o p o r t i o n I 10 20 30 40 50 60 0 . 0 0 0 . 0 3 0 . 0 6 Ag e P r o p o r t i o n y o u 10 20 30 40 50 60 4 . 0 4 . 5 5 . 0 5 . 5 Ag e W o r d l e n g t h 10 20 30 40 50 60 2 0 6 0 1 0 0 Ag e T w e e t l e n g t h 10 20 30 40 50 60 0 . 0 0 . 2 0 . 4 0 . 6 Ag e L i n k s 10 20 30 40 50 60 0 . 0 0 . 2 0 . 4 0 . 6 0 . 8 Ag e H a s h t a g s How old am I?  @USER Woensdagochtend 15 augustus start het landelijke CDA met haar regiotour op Goeree - Overflakkee i.s.m. @USER.  (@USER On Wednesday morning, the 15 th of August starts the national CDA with its tour through the region in Goeree - Overflakkee in collaboration with @USER)  RT @USER: Vanmiddag met @USERMENTION gezellig bij @USERMENTION een wijntje gedaan en naar de Emmauskerk #Middelharnis geweest. Mooie dag zo!  (RT @USER: Had fun this afternoon had wine at @USER with @USER and went to the Emmauschurch #Middelharnis. Beautiful day!) How old am I?  The user is a 19 - year old student  Words related to politics are highly ranked features associated with older users  Prepositions, conventional punctuation, formal abbreviations, conservative vocabulary and for example talking about wine lead to the classification of the user as a 38 - year old employee  People can choose to emphasize aspects of their identity that are not related to gender or age  Deviation from the style and contents from a person ’ s peers: this makes the automatic prediction of gender and age more difficult TweetGenie www.tweetgenie.nl (online soon) TweetGenie: examples  TheoMeder (M, 52)  Antalvdb (M, 43)  Rensbod (M, 48)  Estherouwehand (F, 37)  Claudiadebreij (F, 38)  lalalalinder (Linda Duits, F, 36)  MarcMarieH (M, 48)  GeertWildersPVV (M, 49) vs GeertWilders (fake account) The Twitter Apocalypse  Tweets in totall: 52.848 (retweets: 32.055 = 61%) 0 2000 4000 6000 8000 10000 12000 14000 Number of tweets about 21 - 12 - 2012 NUMBER The Twitter Apocalypse (2)  The most popular retweet (4050 x): “ Those who believe that the world will end on December 21, please donate all your money to my bank account by December 20. ” The Twitter Apocalypse (3)  Belief and religion  LIWC: Linguistic Inquiry and Word Count  Massive fear or media hype? How many Dutch people on Twitter really feared the world would end? The Twitter Apocalypse (4)  2012 - 12 - 16 : @USER: Pam and I are afraid that on December 21 the world will end =' ’ ( The Twitter Apocalypse (4)  2012 - 12 - 16 : @USER: Pam and I are afraid that on December 21 the world will end =' ’ (  Ca. 500 tweets of genuine fear = 1 % Conclusions  Linguistic research / TweetGenie: gender and age  D. Nguyen, R. Gravel, D. Trieschnigg and T. Meder : "How Old Do You Think I Am?" A Study of Language and Age in Twitter at ICWSM 2013 .  Apocalypse: sentiment, popularity, belief  Conduits and micro - variation  Representation of self References  Bamman, David, Jacob Eisenstein and Tyler Schnoebelen. 2012. Gender in Twitter: Styles, Stances and Social Networks. Draft from 23/9/2012.  Barbieri, Federica. 2008. Patterns of Age - based linguistic variation in American English. In: Journal of Sociolinguistics 12/1. 58 - 88.  Boyd, Danah, Scott Golder & Gilad Lotan 2010: ‘ Tweet, Tweet, Retweet: Conversational Aspects of Retweeting on Twitter ’ , in: HICSS - 43. IEEE: Kauai, HI , January 6, 2010.  Naaman, Mor, Jeffrey Boase and Chih - Hui Lai. 2010. Is it Really About Me? Message Content in Social Awareness Streams. CSCW 2010, Savannah, Georgia, February 6 – 10, 2010.  Nguyen, Dong, Noah Smith and Carolyn Rosé. 2011. Author Age Prediction from Text using Linear Regression. Workshop on Language Technology for Cultural Heritage, Social Sciences, and Humanities at ACL.  Nguyen, Dong, Rilana Gravel, Dolf Trieschnigg & Theo Meder: ‘ "How Old Do You Think I Am?" A Study of Language and Age in Twitter ’ , in: ICWSM 2013 (submitted, accepted)  Pennebaker, James and Lori Stone. 2003. Words of wisdom: Language use over the lifespan. In: Journal of Personality and Social Psychology 85. 291 – 301.  Pennebaker, James, Shlomo Argamon and Moshe Koppel. 2007. Mining the Blogosphere: Age, Gender and the Varieties of Self – expression. In: First Monday 12, number 9.  Schler, Jonathan, Moshe Koppel, Shlomo Argamon and James Pennebaker. 2005. Effects of Age and Gender on Blogging. American Association for Artifical Intelligence.