Predicting Elections with Twitter- What 140 Characters Reveal about Political Sentiment
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Predicting Elections with Twitter – What 140 Characters Reveal about Political Sentiment Andranik Tumasjan, Timm O. Sprenger, Philipp G. Sandner, Isabell M. Welpe Workshop „Election Forecasting“ 15 July 2013 Technische Universität München TUM School of Management Lehrstuhl für BWL – Strategie und Organisation Prof. Dr. Isabell M. Welpe
Agenda Introduction and related research Data set and methodology Results and implications Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 2 Prof. Dr. Isabell M. Welpe
The successful use of social media in the last presidential campaigns has established Twitter as an integral part of the political campaign toolbox The increasing use of Twitter as means of …has triggered attempts to better political communication… understand and aggregate this information Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 3 Prof. Dr. Isabell M. Welpe
The goal of our study was to explore 3 research questions Research questions 1 Deliberation Does Twitter provide a platform for political deliberation online? 2 Sentiment How accurately can Twitter inform us about the electorate's political sentiment? 3 Prediction Can Twitter serve as a predictor of the election result? Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 4 Prof. Dr. Isabell M. Welpe
Existing research related to our research questions and resulting research gaps we try to address Research questions Related research Research gap 1 Deliberation Twitter is not only used for one-way Many contexts largely communication, but 31% of all tweets direct a unexplored, e.g. the Does Twitter provide specific addressee (Honeycut & Herring, political debate online a platform for 2009) Unclear whether political deliberation Political internet discussion boards found to findings apply to online? be dominated by a small number of heavy microblogging forums users (Koop & Jansen, 2009) 2 Sentiment 19% of all tweets contain mentions of a brand Limited application to or product and statistically significant political sentiment How accurately can differences of customer sentiment can be Few empirical studies Twitter inform us extracted (Jansen et al., 2009) to explore information about the Pessimism toward the ability of blogs to aggregation in social electorate's political aggregate dispersed bits of information media sentiment? (Sunstein, 2008) 3 Prediction Some studies explore the reflection of the Unclear whether political landscape in "traditional" weblogs findings apply to and social media (e.g., number of Facebook microblogging forums Can Twitter serve as users a valid indicator of electoral success, a predictor of the Williams & Gulati, 2008) election result? Count of candidate mentions in the press can be a better predictor of election results than official election polls (Véronis, 2007) Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 5 Prof. Dr. Isabell M. Welpe
Agenda Introduction and related research Data set and methodology Results and implications Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 6 Prof. Dr. Isabell M. Welpe
We examined more than 100,000 tweets and extracted their sentiment using LIWC Data set Methodology 104,003 political tweets Linguistic Inquiry and Word Count (by James Published between August 13th and September Pennebaker et al.) 19th, 2009 (one week prior to the election) Text analysis software developed to assess Collected all tweets containing the name of emotional, cognitive, and structural either components of text samples using a At least one of the 6 major parties psychometrically validated dictionary Selected prominent politicians Calculates the share of words in a text belonging to empirically defined psychological and structural dimensions LIWC has been used widely in psychology and linguistics including to Measure the sentiment levels in US Senatorial (Yu et al., 2008) Profile politicians Twitter messages Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 7 Prof. Dr. Isabell M. Welpe
Agenda Introduction and related research Data set and methodology Results and implications Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 8 Prof. Dr. Isabell M. Welpe
While Twitter is used as a forum for political deliberation on substantive 1 issues, this forum is dominated by heavy users Two widely accepted indicators of blog-based deliberation… The exchange of substantive issues Equality of participation Party Sample tweet* Users Messages CDU CDU wants strict rules for internet User group Total Share Total Share CSU CSU continues attacks on partner of choice One-time users 7,064 50.3% 7,064 10.2% FDP Light (2-5) 4,625 32.9% 13,353 19.3% FDP Whoever wants civil rights must choose Medium (6-20) 1,820 12.9% 18,191 26.2% FDP! Heavy (21-79) 463 3.3% 15,990 23.1% Grüne After the crisis only Green can help GREEN+ Very heavy (80+) 84 0.6% 14,470 21.2% Total 14,056 100% 69,318 100% SPD Only a matter of time until the SPD dissolves While the distribution of users across user groups Die Linke Society for Human Rights recommends: No is almost identical with the one found on internet government partication for LINKE message boards, we find even less equality of participation for the political debate on Twitter 31% of all messages contain "@"-sign Additional analyses have shown users to exhibit 19% of all messages are retweets a party-bias in the volume and sentiment of their messages * Examples shortened for citation (e.g. omission of hyperlinks) Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 9 Prof. Dr. Isabell M. Welpe
2 The online sentiment in tweets reflects nuanced offline differences between the politicians in our sample LIWC profiles* Leading candidates Other politicians Very similar profile for all leading candidates Positive outweigh negative emotions, except in the Only polarizing political characters, such as liberal case of CSU leader Seehofer who in addition is leader Westerwelle and socialist Lafontaine, associated the most with anger (he irritated many deviate in line with their roles as opposition leaders voters with his attacks on desired coalition partner Messages mentioning Steinmeier, who was FDP) sending mixed signals regarding potential coalition For Steinbrück and zu Guttenberg, the issues partners, are the most tentative money and work, reflect their roles as finance and economics minister * We focused on the 12 dimensions which a priori seemed best suited to profile sentiment and political issues) Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 10 Prof. Dr. Isabell M. Welpe
The similarity of profiles is a plausible reflection of the political proximity 2 between the parties Similarity of LIWC profiles Group Distance* Key findings Politicians All politicians 0.21 High convergence of the leading candidates Governing coalition 0.23 More divergence among Right coalition 0.16 politicians of the governing grand Distance measure to quantify coalition than among those of a the similarity of sentiment Left coalition 0.10 potential right wing coalition profiles The similar profiles of Merkel and Candidates for chancellor 0.02 Steinmeier mirror the consensus- driven style of their grand coalition Leading candidates 0.10 Other candidates 0.24 Parties All parties 0.09 The fit of a potential right-wing coalition is almost as good as the Governing coalition 0.07 fit in the governing coalition Right coalition 0.08 Greatest divergence among parties on the left Left coalition 0.10 Tight fit between sister parties CDU and CSU Union 0.01 * Average distance from the mean profile per category across all 12 dimensions in percentage points Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 11 Prof. Dr. Isabell M. Welpe
The activity on Twitter prior to the election seems to validly reflect the 3 election outcome The share of tweets can be considered a plausible …and joint party mentions accurately reflect the political reflection of the election results… ties between parties All mentions Election results Relative frequency of joint mentions** Vote CDU CSU SPD FDP Linke Party Total Share share Error CDU 30,886 30.1% 29.0% 1.0% CSU 1.25* CSU 5,748 5.6% 6.9% 1.3% SPD 1.23* 0.71* SPD 27,356 26.6% 24.5% 2.2% FDP 1.04* 1.01 0.90* FDP 17,737 17.3% 15.5% 1.7% Die Linke 0.81* 0.79* 1.04* 0.97 Die Linke 12,689 12.4% 12.7% 0.3% Grüne 0.84* 0.79* 0.98 1.06* 1.18* Grüne 8,250 8.0% 11.4% 3.3% MAE = 1.65% Research institute MAE (last poll) Forsa 0.84% An analysis of messages surrounding the Forschungsgruppe Wahlen 1.04% TV debate between the main candidates GMS 1.48% has shown that tweets can also reflect the sentiment over time Infratest/dimap 1.40% * Significant at the .05-level ** Measures how often two parties are mentioned together relative to the random probability Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 12 Prof. Dr. Isabell M. Welpe
Our findings suggest the use of social media information content to complement insights regarding the public's political sentiment Research questions Conclusions 1 Deliberation While we find evidence of a lively political debate on Does Twitter Twitter, this discussion is dominated by a small provide a platform number of users: only 4% of all users account for for political deliberation more than 40% of the messages online? 2 Sentiment How accurately Sentiment profiles plausibly reflect many nuances of can Twitter inform the election campaign us about the Politicians evoke a more diverse set of profiles than electorate's parties political Similarity of profiles is indicative of the parties' sentiment? proximity with respect to political issues 3 Prediction In contrast with previous studies of political message Can Twitter serve boards, we find that the mere number of messages as a predictor of reflects the election results and even comes close to the election traditional election polls result? Joint party mentions mirror closeness on political issues and likely coalitions Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 13 Prof. Dr. Isabell M. Welpe
Summary and discussion Aftermath Open questions and challenges Currently ~ 350 citations (since 2010) Sampling time frame Several attempts to replicate or Constantly changing user number and “extend”/“enhance” our approach in other demographics in Twitter electoral contexts Type of mentions (candidates, party, …) Countries Keyword selection (full names, Time intervals abbreviations…) Election types (e.g, primaries) Type of analysis (simple counts, sentiment, Constituencies (e.g., counties) algorithms, input data…) Mention types (e.g., candiates) Type of elections (primaries, parliament, …), Analytical methods (e.g., senitment) constituencies, and political systems Preliminary result of own literature survey Trustworthiness of tweets (depending on aspiration level) File drawer problem 11 rather positive papers Aspiration level (replace or complement 7 rather negative papers other forecasting methods) National level results tend to be more “Real” replications hardly possible supporting of our initial findings than other … election types Longer time frames more accurate Party mentions tend to be more accurate than candidate mentions Partly based on Gayo-Avello (2012) Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 14 Prof. Dr. Isabell M. Welpe
hank you for your attention! Technische Universität München Lehrstuhl für BWL - Strategie und Organisation 15 Prof. Dr. Isabell M. Welpe
You can also read