Limits of Electoral Predictions Daniel Gayo-Avello (@PFCdgayo) – Univ. of Oviedo (Spain) Panagiotis T. Metaxas (@takis_metaxas) – Wellesley College (USA) using Twitter Eni Mustafaraj (@enimust) – Wellesley College (USA) Political discourse in social media is becoming common practice. One interesting aspect of this is the possibility of pulsing the publics opinion about the elections. Allegedly, predicting Promising results? 7.6% is a far cry from 2-3% MAE. Sentiment analysis is just slightly better NO! electoral outcomes from social media data can be feasible a even simple. Positive results have than random classifier (36.9%). been reported, but without an analysis on what principle enables them. Our work puts to test the purported predictive power of social media metrics against the 2010 US congressional Sentiment analysis weakly correlates with elections. We found no correlation between the analysis results and the electoral outcomes, user’s political preference. contradicting previous reports. We argue that one should not be accepting predictions about events using social media data as a black box. Instead, scholarly research should be accompanied by a model explaining the predictive power of social media, if there is one. Social media users are not a representative unbiased sample of likely voters. 22% of adult users engaged in Social media data is easily manipulated by electoral campaigns Unpredictability spammers and propagandists. through OSNs in 2010. Motivation Social media data have been used to of elections with (In Twitter nobody knows you are a robot). successfully “predict the present” in social media “It works” is not enough: we need to know a great variety of topics. how it works. (Much) better (and subtle) sentiment Can social media data predict elections? analysis methods are needed. YES WE CAN NO WE CANNOT Maybe. Who knows? Williams & Gulati Goldstein & Rainey (L.A. Times 2010) O’Connor et al. (ICWSM, 2010)(Politics and Technology Review, 2008) Jungherr, Jürgens & Schoen Lui, Metaxas & Mustafaraj Carr (Fast Company, 2010) Tumasjan et al. (ICWSM, 2010) (Social Science Computer Review, 2011) (e-Society Conference, 2011) Gayo-Avello (CACM, 2011) Therefore… <flame> Can we replicate the results? Social media data might predict but it is difficult. everything 17.1% (Twitter volume) Mean Absolute Error 7.6% (Twitter sentiment) Corollary: Social media data can predict everything (MAE) 2-3% (professional polls) once the answer is known and data not fitting are ignored. Winner predicted in only half of the races. </flame>
A particular slide catching your eye?
Clipping is a handy way to collect important slides you want to go back to later.