In this study, we demonstrate that the computational social science is important to understand people behavior in political phenomena, and based on the long-running Brexit debate analysis on Twitter, we predict the public stance, discussion topics, and we measure the involvement of automated accounts and politicians’ social media accounts.
Analysis of On-line Debate on Long-Running Political Phenomena.The Brexit Case
1. Analysis of On-line Debate
on Long-Running Political Phenomena.
The Brexit Case
Marco Brambilla, Emre Calisir
@marcobrambi
Politecnico di Milano, Data Science Lab
Amsterdam
July 18 2019
2. The New Agora
• Social media platforms are the new places for political debate
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
3. Political Discussions on Social Media
Not only drama (uprisings, riots, ..)
◢US Presidency (Obama, Trump)
◢National elections throughout the world
◢Referendums (Catalunya, Lombardia, Brexit)
◢Political agenda (immigration, education, ..)
What can you find there?
◢ Direct debates between candidates
◢ Criticisms and defense of political actions
◢ Discussions between voters
◢Echo of events and facts
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
4. A Comprehensive Approach to the Phenomenon
◢Key Events
◢Topics
◢Main Actors
◢Audience opinion
◢Sentiment
◢Spreading of info
◢Robotization
◢Dynamic aspects
Stance
Topic
Discovery
Top
Influencers
Temporal
Analysis
Demographics Sentiment
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
5. BREXIT
Research Questions
• How can we analyze polarized political
elections such as referendums with social media
data?
• How public mood is changed before and after
the referendum?
• What is the influence of politicians to the online
debate?
• What is the influence of automated accounts to
the online debate?
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
6. Twitter. Not just because …
Statistics
◢800 mln monthly active users
◢500 mln daily active users
Features
◢People tend to share political opinions
◢Information spreading speed
◢Public tweets
Text content
Mentions, hashtags, retweets, likes
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
7. Twitter. Not just because …
10M+ tweets containing Brexit keyword
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
8. Analysis setting
• Collection of tweets containing the keyword Brexit
• neutrality of the term proven by empirical studies
• between January 2016 and October 2018 (.. But ongoing)
• using Twitter API
• 10 million tweets sent by 1.5 million users
• multi-language
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
10. User Participation to Twitter Debate
• Limited attention span
56% of users tweeted using Brexit keyword only once in the 28 months of our analysis
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
11. The Language of Brexit
81% of tweets are posted in English.
Tweets written in French and Spanish, Italian and German around 2-4%
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
12. Geo-Location of Brexit Tweets
There is an interest from all over the world, while
not surprisingly most of them are posted from UK.
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
Basedongeotaggedposts
around 45% from the UK
13. Demographics: Gender of users
Gender distribution is very similar to Twitter population
Infoextractedfromprofilephotos(around30%ofdata)
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
14. Age of users
Users who discuss Brexit are older than Twitter distribution.
Scarce attention from young people.
Infoextractedfromprofilephotos(around30%ofdata)
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
15. Temporal Analysis to detect key events
Daily unique users participating to Brexit debate
1
2,3
4,5
6,7,8
109
Vote
BrexitProcess
start
Trum
p
elected
M
ay’s
speechBrexitBill&
Article
50
N
egotiations
start
Firstdealfails
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
16. Daily Increases > 150%
(Daily)
Mapping with Key Event
1 June 24, 2016 UK votes to leave EU
2 Nov 3, 2016 Parliament must vote on whether the UK can start the process of leaving the EU, the
High Court has ruled.
3 Nov 9, 2016 Donald Trump is selected as US president, and European commission congratulated
Trump’s victory.
4 Jan 17, 2017 Theresa May’s Brexit speech
5 Jan 24, 2017 The Supreme Court is ruling Brexit delivery
6 Mar 13, 2017 Parliament passes Brexit bill and opens way to triggering article 50
7 Mar 29, 2017 Prime Minister triggers Article 50 of the Treaty on European Union
8 Apr 18, 2017 Prime Minister calls a General Election – to be held on 8 June 2017.
9 June 19, 2017 UK-EU exit negotiations start
10 Dec 4, 2017 UK and EU fail to strike Brexit talks deal
Corresponding Events
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
17. Recent Brexit tweets are more influential
Avg favorites
Avg retweets
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
20. User Stance evolution in time
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
Based on use of polarized hashtags
21. Supervised Learning Pipeline
for Stance Classification
Human-in-the-loop paradigm
Training from human-annotated and rule-based training
Training Dataset
Balanced tweets from
Remain Leave and
Neutral
Feature Engineering
+ Transformations
Learning Model
SVM, Log Regr. Random
Forest, LSTM
Evaluation
Adjustments
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
22. Increasing engagement of pro-remain right after the
referendum
Increase in
Pro-Remain
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
23. Increasing engagement of pro-leave later
Increase in
Pro-Leave
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
24. Stance in time (%) on the whole dataset
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
25. Another Perspective: Is there any change in stance
after the referendum?
62% of Pro-Leavers moved to Pro-Remain stance after the referendum
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
26. Top influencers are the accounts that
other users discussed about
Politicians News Channels Campaign Accounts
@UKLabour 44K
@Conservatives 31K
@LeaveEUOfficial 25K
@vote_leave 24K
@UKIP 18K
@LibDems 16K
@StrongerIn 16K
@theresa_may 68K
@Nigel_Farage 53K
@jeremycorbyn 44K
@BorisJohnson 41K
@David_Cameron 33K
@realDonaldTrump 30K
@DavidDavisMP 19K
@NicolaSturgeon 11K
@JunckerEU 11K
@MichelBarnier 10K
@ChukaUmunna 10K
@BBCNews 39K
@SkyNews 28K
@guardian 28K
@FT 27K
@LBC 22K
@Independent 15K
@Telegraph 13K
@afneil 11K
@MailOnline 10K
Top mentioned accounts
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
27. Temporal and comparative analysis of Brexit
tweets corresponding to politician names
Handing over PM from Cameron to Theresa, the visit of Trump...
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
28. Adding sentiment…
• Stance and sentiment (may be) orthogonal
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
29. Analysis of top mentioned politician accounts in
stance and sentiment dimensions
Politician who is discussed most positively:
Donald Trump and Nigel Farage
Politician who is discussed most negatively:
David Cameron and Jeremy Corbyn
Pro-remainers discuss more about Boris
Johnson
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
30. Topic Analysis + User Stance + Tweet Sentiment
(2018) [part 1]
Topic Stance Sentiment Representative Words
31. Topic Analysis + User Stance + Tweet Sentiment
(2018) [part 2]
Topic Stance Sentiment Representative Words
33. The higher the bot score,
the more likely to have Pro-Leave stance
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
35. Conclusions and Future Work
PRESENT
• A cool use case! Some surpri
• A general method and implementation, configurable
• Full paper https://arxiv.org/abs/1901.00740
FUTURE
• Continuous collection of data
• Analyze evolution of topics (topic drifting)
• Apply Graph-based models
• Deepen the social science interpretation part
• Resources published (open data + code)
Brambilla, Calisir. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case.
36. Analysis of On-line Debate on Long-Running Political Phenomena. The Brexit Case
Thanks! Questions?
@marcobrambi
Data Science Lab
Politecnico di Milano, Italy
marco.brambilla@polimi.it
Politecnico di Milano, Italy
http://datascience.deib.polimi.it/
@datascience_mi