Upload
social-physicist
View
365
Download
1
Tags:
Embed Size (px)
Citation preview
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
Using Network Science to Understand Elections:The 2014 South African National Elections on Twitter
Kyle Findlay & Ockert Janse van Rensburg
Winner of the Gold Award for Best Paper at the 2015 Southern African Marketing Research Association (SAMRA) annual conference
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
2
Let’s start by having a look at the data in action…
(video animation on next slide)
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
In the interest of privacy, we do not report on all influencers in this paper. We stick to only reporting those that already have a significant public presence, either in the South African media (e.g. politicians) or on Twitter itself (i.e. more than 10,000 followers).
This data is from Q2 2014. A lot can change in a year, including people’s opinions and political allegiances. Again, community membership should not be taken alone as proof of political views nor allegiances.
A few important caveats before we begin…
3
This paper uses network theory-based approaches to identify communities of Twitter users within the data. It is important that readers understand that community membership does not 100% identify nor guarantee a user’s political views nor alignment. There are two main reasons for this:
1. Community membership is based on a non-deterministic algorithm that uses a random seed to start the community detection process. This means that community membership can be unstable and so reported memberships should be taken with a pinch of salt.
2. In simplistic terms, users are grouped into communities based on who they interact with most. Generally speaking, people tend to interact with other like-minded people; however, antagonistic interactions can also bind communities i.e. people may form part of the same community due to debates wherein users engage each other on their differing viewpoints.
Community membership
Privacy Time period
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
Contents
4
1. So what were the conclusions?
2. Why focus on Twitter?
3. How did we analyse the data?
4. Party momentum
5. Overall election influencers & top content
6. The SA elections 2014 conversation map
7. Democratic Alliance (DA) community
8. Economic Freedom Fighters (EFF) community
9. Disenchanted ANC (ANC) community
10.ANC Stalwarts (ANC) community
So what were the conclusions?
5
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
President Jacob Zuma has split the ANC in two
6
Many supporters happily engage with official party mouthpieces such as @MyANC_, Sports Minister Fikile Mbalula and @ANC_Youth.
However, many disenchanted millennials appear to have found their thoughts echoed by unofficial influencers such as Khaya Dlanga, @Mtshwete and @TaxiDriverSipho
Support for the ANC is unequivocal amongst these groups but half of the party’s supporters do so despite of their dissatisfaction with Jacob Zuma
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
The 2014 national elections conversation on Twitter consisted of four main constituencies
7
DA DA
EFF EFF
Dis-enchanted
ANC
Dis-enchanted
ANC
ANCstalwarts
ANCstalwarts
…which encompassed 52% of all users in the conversation
…and generated 85% of all tweets!
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
EFF community
@City_Press
@POWER987News
@SABCNewsOnline
@SAfmnews
@Radio702
ANC stalwarts
@ANN7tv
@The_New_Age
@SAgovnews
Some news media outlets’ content resonated primarily with specific constituencies
8
DA community
@RSApolitics
@JacaNews
@BDliveSA
@dailymaverick
Disenchanted ANC
[No news entity appeared to cater specifically to, nor particularly resonated with, this community]
Many news outlets formed their own independent communities. For example…
• eNCA
• EWN News
• Mail & Guardian
• News24
• Times Live
• SA Breaking News
However, these news outlets’ content primarily resonated within the following communities:
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
The distribution of Twitter mentions gives us tantalising clues at possible future party momentum
9
62%
22%
6%3% 2% 1% 1% 1% 1% 1% 1% 0% 0%
55%
22%
14%
1% 0% 1% 1% 2%4%
ANC DA EFF IFP NFP FF+ UDM ACDP AIC COPE Agang APC PAC
Seats in parliament % Twitter mentions
If the entire country was on Twitter, would the election results have looked like this?
Why focus on Twitter?
10
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
Only 13% of South Africans belong to Twitter
11
100%
46% 42%
13% 10% 7% 2%
Total consumermarket
Total socialnetworking
users
Facebook Twitter Google+ YouTube Instagram
% belong to
Source: TNS Sunday Times Top Brands Survey 2014
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
12
“BuzzFeed found that
Twitter has a big cascade effect on other social media platforms.
Put simply, it appears that
huge stories often start as tweets,
then get shared by influencers to Facebook and other networks, where the original piece of content subsequently gets far more distribution.”
…however, Twitter has an outsized effect on information spread
Source: http://www.fastcompany.com/3043788/sxsw/twitters-influence-problem-visualized
How did we analyse the data?
13
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
14
user location
1,461,909(3 March – 12 May 2014)
user time-zoneuser languagetweet language
We started with 1.5m tweets about the elections which we cleaned extensively to remove irrelevant tweets…
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
15
981,878tweets in the end
irrelevant hashtagsirrelevant influencersirrelevant retweetsk-means clusteringconversation network
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
16
We connected users that interacted with each other
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
17
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
18
…and then ran a community detection algorithm to identify distinct communities
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
We ‘clustered’ tweets into topics using Latent DirichletAllocation (LDA)
19
Party momentum
20
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
21
248
89
25
10 6 4 4 3 3 3 2 1 1
ANC DA EFF IFP NFP FF+ UDM ACDP AIC COPE Agang APC PAC
Seats in parliament
A reminder of the actual election results…*
* …which follow a power law (R² = 0.97)
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
22
62%
22%
6%
3% 2% 1% 1% 1% 1% 1% 1% 0% 0%
38%
25%
13%
3% 2% 1% 1% 1% 0%
4% 3%0% 1%
ANC DA EFF IFP NFP FF+ UDM ACDP AIC COPE Agang APC PAC
% of seats in parliament % of media coverage
R=0.95
Party media coverage aligned fairly closely with the actual results
Source: Media Monitoring Africa Election Coverage 2014, http://elections2014.mediamonitoringafrica.org
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
23
62%
22%
6%
3% 2% 1% 1% 1% 1% 1% 1% 0% 0%
55%
22%
14%
1% 0% 1% 1% 2%4%
ANC DA EFF IFP NFP FF+ UDM ACDP AIC COPE Agang APC PAC
% of seats in parliament % Twitter mentions
R=0.99
…but each party’s share of Twitter mentions aligned even better with their actual results
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
24
Twitter share of mentions notably diverged from actual seats in parliament in two cases…
The ANC received less than its fair share of Twitter mentions than we would expect given its final number of parliamentary seats won
The EFF received more Twitter mentions that seats won The DA received
exactly its fair share of mentions versus seats won
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
25
The results might imply that, if South African Twitter users were more representative of the voting public,…
…the ANC might have received far fewerseats (62% vs. 55%)
…and most of these losses might have come from the EFF which might have received a greatershare of seats (6% vs. 15%)
…while the DA’s voter block probably is more representative of its Twitter users, thus its share of seats neatly aligns with its share of Twitter mentions (22%)
What might these results imply (with a very healthy dose of speculation)?
However, it’s important to remember that this interpretation is
very speculative at best!
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
26
The scientific debate on the predictability of Twitter mentions vs. elections is still undecided though…
Source: Young-Ho, E, et al. (2015). Twitter-based analysis of the dynamics of collective attention to political parties
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
27
“Despite the many efforts, results are still inconclusive...”
“We conclude that the tweet volume is a good indicator of parties' success in the elections when considered over an optimal time window.”
Source: Young-Ho, E, et al. (2015). Twitter-based analysis of the dynamics of collective attention to political parties
Overall election influencers & top content
28
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
29
39 008
34 592
31 481
27 746
12 855
12 764
10 478
10 051
9 602
8 085
7 671
7 629
6 649
6 379
6 364
6 050
5 092
5 031
4 735
4 639
@HelenZille
@MyANC_
@DA_News
@EconfreedomZA
@ANC_Youth
@Julius_S_Malema
@Maimanea
@MbalulaFikile
@AgangSA
@eNCANews
@News24
@LindiMazibuko
@City_Press
@SABreakingNews
@POWER987News
@Sentletse
@EWNReporter
@ChesterMissing
@SAPresident
@TimesLive
Top 20 influencers shown where influence = # interactions (retweets + mentions)
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
30
@HelenZille
@MyANC_
@DA_News
@EconfreedomZA
@ANC_Youth
@Julius_S_Malema
@Maimanea
@MbalulaFikile
@AgangSA
@eNCANews
@News24
@LindiMazibuko
@City_Press
@SABreakingNews
@POWER987News
@Sentletse
@EWNReporter
@ChesterMissing
@SAPresident
@TimesLive
Democratic Alliance
African National Congress
Economic Freedom Fighters
News media
Top 20 influencers shown where influence = # interactions (retweets + mentions)
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
31
10
6
18 15
21
20
16
29
The top hashtags give us some insight into the most prevalent topics (see hashtags highlighted with arrows)
Top 30 hashtags. Excludes “elections”- and “South Africa”-related hashtagsNumbers inside arrows represent rank of hashtag in term of frequency of occurrence in data
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
32
What was the top 20 most retweeted content?
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
Author Retweet content # retweets
@justicemalalaMbeki's #ANC in 99: 66.35% Mbeki's ANC 2004: 69.6% Zuma's ANC 2009: 65.9% Zuma's ANC 2014: 62.84
(22.12/08 May) #justsay…705
@TrevornoahJacob Zuma is a great. I bet he did this Nkandla thing just to unite all South Africans in a common anger at
corruption.590
@ProudlySACongratulations to Deputy President Motlanthe & his beautiful bride, Gugu on their wedding today!! @PresidencyZA
http://t.co…524
@IECSouthAfricaNational Assembly seats: APC–1; PAC–1; AGANG SA–2; ACDP–3; AIC–3; COPE –3; UDM–4; VF Plus–4; NFP–6;
IFP–10; EFF–25; DA–…490
@helenzilleBy saying that "only clever people" have problems with R246-mill Nkandla upgrade, Pres Zuma is implying that
ANC voters are…394
@[] @IECSouthAfrica apparently found in home of ANC party agent. https://t.co/OyoYIvI79Y 346
@alexeliseevZuma on #Nkandla: "It's not an issue with voters. It's an issue with bright people. Very clever people". What is he
saying…344
@Sentletse The ANC and the IEC must be ashamed of themselves! http://t.co/STwypvLg8M 342
@IECSouthAfrica I.E.C voting material found at a house of an anc party agent in ward 77 http://t.co/48MAjhqqXE" 326
@[]EFF votes found dumped near Diepsloot cc @Julius_S_Malema @Sentletse @EconFreedomZA with @IECSouthAfrica
stamp http://t.co/mIzvrow…325
@Julius_S_Malema We will soon announce the date of the march to the union buildings to demand Zuma's resignation as the president 294
@helenzilleMonitor the polls!! "@E_van_Zyl_17: "@JohnBiskado: IEC voting material found at house of anc party agent ward
77 http://t.…289
@MbalulaFikile But you cant use the same BIS bought with NSFAS to tweet "ANC HAS DONE NOTHING". Uxokelani? 280
@GarethCliff Oh no! RT @DonnyDunn: @GarethCliff @KienoKammies IEC integrity obliterated! http://t.co/wILOizXGXm 279
@MbalulaFikile Biggest loser of the century DR Mamphela Ramphela ,u Luzile shameeee 272
@TannieEvitaConfusius say: "He who knows nothing about his own house, knows even less about his own country."
@SAPresident271
@GarethCliffWho told @helenzille it would be a good idea to do this? Will she campaign in blackface next?
http://t.co/TzbwsqgHGL258
@GarethCliffEven if you didn't vote ANC, they will form your new govt. wishing them luck is wishing the best for all of us. Good
luck …268
@MaxduPreezDefend this, ANC: Public Works Dept redirected service delivery funds to pay for Nkandla in contravention of
constitution245
@MbalulaFikileYet you're on Twitter and can write that in perfect English, let's face it, you're ANC's good story. @lusylooya: ANC
ha…238
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
34
And if we group the retweets by theme?
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
Ballot tampering
Nkandla
Political smack talkFikile Mbalula and the ANC’s ‘good story’
Election results & well wishingDeputy President Motlanthe‘s wedding
Malema march for Zuma’s resignation
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
Many of the top most shared images related to alleged ballot tampering
36
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
37
This image drawing a parallel between President Jacob Zuma’s Nkandla scandal to e-tolling was one of the most popular
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
38
…as was this image critical of DA leader, Helen Zille
The SA elections 2014 conversation map
39
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
The overall elections conversation map
40
The various regions of colour clearly highlight distinct communities in the elections conversation.
Specific influential accounts that led the conversation are clearly visible within each community, hinting at the agenda of each.
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
41
Black influencers
?
EFF
DA(and FF+)
ANC
eNCA & EWN News
International news media
Gareth Cliff & Ulrich J van Vuuren
Comedians
Agang
News24 & TimesLive
SA Breaking News
IEC & PresidencyZA
Mail & Guardian
DJ Sbu
The top four communities encompass 52% of unique users and appear to relate to political parties. Other communities relate to news entities, DJs, comedians, etc.
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
42
Black influencers
?
EFF
DA(and FF+)
ANC
However, the top four communities generated almost all of the tweets about the elections (85%)!
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
43
It was not initially clear what the “black influencers” community stood for. It required some digging into their actual tweet topics to find out more…
Black influencers
???
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
44
…so let’s unpack the top four communities in more detail…
Democratic Alliance (DA) community
45
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
46
Mabine Seabe II@Mabine_Seabe
Lindiwe Mazibuko@LindiMazibuko
Helen Zille@helenzille
Democratic Alliance@DA_News
Mmusi Maimane@MaimaneAM*
RSApolitics@RSApolitics
Jacaranda News@JacaNews
ToxiNews@toxinews
Gavin Davis@gavdavis
Influencers
Top influencers ranked on # interactions (retweets + mentions). * Username has changed subsequently. Original account hacked?
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
47
BDlive@BDliveSA
RSApolitics@RSApolitics
Jacaranda News@JacaNews
ToxiNews@toxinews
Daily Maverick@dailymaverick
Particularly resonant media entities
These news media accounts’ content resonated with this community more than any other (most mentions and retweets)
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
48
Nkandla & Zuma corruption
Ballot tampering
Zille pro (e.g. made the party)
Zille against (e.g. Twitter meltdown)
Mazibuko for president
Maimane for president
ANC “good story” (sarcastic)
Topics of conversation
Summary based on top retweeted content and LDA topic models
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
49
Top shared media
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
50
Top shared media
Economic Freedom Fighters (EFF) community
51
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
Redi Tlhabi@RediTlhabi
POWER987 News@POWER987News
EFF Official Account@EconFreedomZA
Julius Sello Malema@Julius_S_Malema
City Press Online@City_Press
Sentletse@Sentletse
SABC News Online@SABCNewsOnline
Ranjeni Munusamy@RanjeniM
Carien du Plessis@carienduplessis
52
Influencers
Top influencers ranked on # interactions (retweets + mentions).
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
53
SAfm news@SAfmnews
702@Radio702
City Press Online@City_Press
POWER987 News@POWER987News
SABC News Online@SABCNewsOnline
Particularly resonant media entities
These news media accounts’ content resonated with this community more than any other (most mentions and retweets)
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
54
Ballot tampering
March for Zuma’s resignation
ANC decline under Zuma
ANC’s “good story” (sarcastic)
Armed Bekkersdal ANC supporter
Topics of conversation
Summary based on top retweeted content and LDA topic models
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
55
Top shared media
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
56
Top shared media
Black influencers (???) community
57
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
58
Trev@Tokyo_Trev
Siya@Siya_THATguy*
Khaya Dlanga@khayadlanga
Mayihlome@MTshwete
IG: TaxiDriverSipho@TaxiDriverSipho
Nzinga@NzingaQ
I.G: Questionnier@Questionnier
L’Vovo Derrango@LvovoSA
DJ Lulo Cafe@LuloCafe
Influencers
Top influencers ranked on # interactions (retweets + mentions). NOTE: some users with <10k followers not shown for privacy reasons* Username has changed subsequently. Original account hacked?
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
59
N/A
No media entities resonated primarily with just this community
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
60
Long live the ANC
ANC’s good story
ANC’s decline under Zuma
No respect for Zuma
Voting ANC (some voting DA)
Congrats to Lindiwe Mazibuko
Zille ‘trying too hard’
Exasperation with EFF (and some support)
When we unpacked their topics of conversation, it became clear what defined them – their disappointment in Jacob Zuma!
Summary based on top retweeted content and LDA topic models
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
61
Top shared media
African National Congress (ANC) community
62
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
63
ANN7 24-hour news@ANN7tv
Malusi Gigaba@mgigaba
ANC Info Feed@MyANC_
ANC Youth League@ANC_Youth*
Fikile Mbalula@MbalulaFikile
ANC-HISTORY@ANC_LECTURES
Vote @MyANC_!!!!!@anccadres
ANC Gauteng@GautengANC
The New Age@The_New_Age
Influencers
Top influencers ranked on # interactions (retweets + mentions). NOTE: some users with <10k followers not shown for privacy reasons* Username has changed subsequently. Original account hacked?
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
64
ANN7 24-hour news@ANN7tv
The New Age@The_New_Age
SA Gov News@SAgovnews
Particularly resonant media entities
These news media accounts’ content resonated with this community more than any other (most mentions and retweets)
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
65
Love ANC
Love Zuma
ANC’s “good story”
Topics of conversation
Summary based on top retweeted content and LDA topic models
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
66Summary based on top retweeted content and topic models
No specific media was particularly popular within this community within our data
Kyle Findlay & Ockert Janse van RensburgUsing Network Science to Understand Elections: The 2014 South African National Elections on Twitter
Thank you