ICWSM2023 The Chance of Winning Election Impacts on Social Media.

Presentation at ICWSM'23

Transcript

1. The Chance of Winning Election Impacts on Social Media Strategy Taichi Murayama1, Akira Matsui2, Kunihiro Miyazaki2, Yasuko Matsubara1, Yasusi Sakurai1 1 SANKEN, Osaka University 2 Yokohama National University 3 Indiana University Bloomington ICWSM 2023
2. 2 Activation of social media use for election Background More people get their news from social platforms Politicians frequently use social media for advertising
3. 3 Rise of fringe candidates in elections Background U.S Increasing of the number of presidential election candidates from parties other than the two major parties Japan Trumpian-inspired party (Sansei-to) wins election
4. 4 Research Question RQ: What is the differences in candidates’ social media strategies during elections in response to the chances of winning? Candidate A Candidate B Almost Win! Almost Lose! Candidate A @CandidateA Candidate A @CandidateA Look it ! Hey! Candidate B @CandidateB Candidate B @CandidateB What is this? Nya! What is the differnce between their strategy?
5. 5 Research Question RQ: What is the differences in candidates’ social media strategies during elections in response to the chances of winning? Unraveling this RQ from four perspectives 1. User behaviors based on basic statistics 2. Content 3. Relationship between content and user engagement 4. Communication
6. 6 l Target: Candidates in 2022 Upper House of Councilors Election in Japan l Platform: Twitter %BUBTFU l The chance of winning is based on Asahi Shinbun’s pre-election prediction
7. 7 l We devided the candidates’ accounts on Twitter into four groups l W: almost Winning l E: even l L: almost Losing l PR: Proportional Representation %BUBTFU Constituency system in each prefecture Two election ways
8. 8 Comparison between each group: l Number of following, followers l Number of tweets l Account Age 1: The characteristics of user behaviors
9. 9 Comparison between each group: l Number of following, followers l Number of tweets l Account Age 1: The characteristics of user behaviors The majority of users only follow 100 – 1,000 users Following
10. 10 Comparison between each group: l Number of following, followers l Number of tweets l Account Age 1: The characteristics of user behaviors • Group W > Group L • More followers does not necessarily increase the chance of winning an election Follower
11. 11 Comparison between each group: l Number of following, followers l Number of tweets l Account Age 1: The characteristics of user behaviors Group L is younger than other groups Account Age
12. 12 1: The characteristics of user behaviors ・Group PR is most tweet, and retweets ・ Group L is most replies. ・ Group W < Group L During Election
13. 13 2: Content l Understanding the topics to which each group tends to refer l Biterm topic model to all tweets in each group Biterm Topic Model 28 Topics Schedule Tax COVID-19 Streaming
14. 14 2: Content l Understanding the topics to which each group tends to refer l Biterm topic model to all tweets in each group Topics with a largen and small of tweets • During pre-election, Schedule, Diplomacy, Please, and Campaign are popular topic • During election, they focus on the promotion of themselves. Otherwise, they do not focus on political discussions, such as Childcare, Income and Constitution.
15. 15 2: Content l Understanding the topics to which each group tends to refer l Biterm topic model to all tweets in each group Topics with a largen and small of tweets • During pre-election, Schedule, Diplomacy, Please, and Campaign are popular topic • During election, they focus on the promotion of themselves. Otherwise, they do not focus on political discussions, such as Childcare, Income and Constitution.
16. 16 2: Content l Understanding the topics to which each group tends to refer l Biterm topic model to all tweets in each group Topics with a large tweets compared to other groups • Group W focuses on Free topic • Group E posts tweets related to the area in which they are running for office • Group L posts Please tweets, urging people to vote and support the campaign.
17. 17 3: Relationship between content and user engagement l We explore what type of content is likely to gain user engagement in each group? l Using a linear mixed-effects model 28 Topics Schedule Tax COVID-19 Streaming Candidate A @CandidateA We should introduce a consumption tax and invest more money in state projects! Candidate A 100 Tax Group L To what extent do Group L's Topic Tax posts contribute to user engagement? 100
18. 18 3: Relationship between content and user engagement l Findings Overall Trends l Topics regarding political policies (Tax, Economy, Childcare) tend to get more retweet l Campaign (the topic of campaign speech) and Schedule (the topic of reporting upcoming events) are not likely to get a lot of retweets. l Group E tends to get more retweets for Soapbox (the topic of their own soapbox oratory) and PR (the topic of the support for proportional representation) l Group L tweets about Please topic, but these tweets can not get more retweets.
19. 19 4: Communication l What is the differences in replying to others for each group? Group W Group L Reply What’s difference? Reply Other users
20. 20 4: Communication l What is the differences in replying to others for each group? Group W Group L Reply What’s difference? Reply Other users Group L and PR have more users replied than Group W and E. Number of reply target users during Election term
21. 21 4: Communication l What is the differences in replying to others for each group? Candidates in groups W and E focus more on conversations with users with a verified badge and with many followers than general users. 40% of all reply targets are verification accounts in Group W, while 18% of those are verification accounts in Group L. Number of followers in the reply targets 0.683 0.751
22. 22 4: Communication l What is the differences in replying to others for each group? • During election term, average sentiment score is the order as PR (0.897) > L > W > E (0.820) • Candidates who are more likely to be elected have fewer positive replies than other groups. Time series of sentiment in the reply targets 0.683 0.751
23. 23 l We finds that the attitude in political communication and the topics they talk about differ depending on the chance of winning an election. l Candidates who are likely to lose the election (Group L) make a lot of posts asking for votes, but they don't get those engagements. l Candidates who are expected to run in close races (Group E) will increase the number of posts about the area in which they are running. l As their chances of winning increase (Group W), candidates narrow the targets they communicate with, from people in general to the electoral districts and specific persons. Conclusion