The First International Workshop on News and Public Opinion 17 May 2016 Cologne, Germany Different News for Different Views Political News-sharing Communities on Social Media Through the UK General Election in 2015 Matt J. Williams Biosciences University of Exeter Iulia Cioroianu Politics University of Exeter Hywel T. P. Williams Biosciences University of Exeter @voxmjw
2005 An evolving media environment... 2010 2015 5.4M* 7M UK 15M UK 28M UK UK’s first ‘social media election’ 31M UK The ‘hashtag election’ – Mark Frankel, BBC, A.Ed. of Social News
Social Media Exposure ‘Legacy’ Media Exposure Sources: Multiple online platforms Each platform 10s-100s of connections Communication exposure [share + re-share] Inherently networked Hard to audit? Bias: Bias and exposure are an emergent property Diverge range of exposure profiles throughout network Sources: <= 10 sources? Selected by reader Easy(ish) to audit Bias: Selection bias Editorial bias print, digital news, radio, television, ... Twitter, Facebook, Instagram, YouTube, ... ? ? ? ? ? ? ? ? ? Williams (2015) Glob. Env. Change, 32, 126-138
Climate change activists vs skeptics Williams (2015) Glob. Env. Change, 32, 126-138 Twitter climate change followers network Social media utopia? open marketplace for ideas; global debate and consensus Social media dystopia? filter bubbles; echo chambers; segregation; polarisation
This study... Exploring political news exposure and the interaction between news and social media during the UK general election Datasets... Tools... Election news content: 14k articles; 17 UK newspapers Social media [Twitter]: 87k tweets; 52k users; 11k hashtags Automated content analysis Network analysis How did GE discussion evolve in news media and social media? What GE-focused news-sharing communities exist? What are their idealogical biases? How does their media exposure differ? Questions... Ongoing work Better/bigger data! Genpop exposure vs politicians’ exposure
Dataset: News articles 21 topics moderate right moderate left centre 6 political party topics 15 issue topics NHS(Health) Tax&Spend Polls Debates Housing London Coalition EU Business Regions Schools Benefits Media Economy Donations 17 national and local papers 13,551 election news articles 22 Mar – 17 May
Task: Identifying election-related news articles LexisNexis News Articles Database All articles across 17 UK newspapers over 4 months (100k-450k articles) Human coding 11,000 article training set. Articles classified as GE and not-GE. ~10 human annotators Supervised classifier (SVM) Cross-validation: F-score = 0.95 13,551 articles classified as General Election related
GE news articles corpus Topic modelling (latent Dirichlet allocation) 13,551 election-related news articles each article gets tagged with mixture of 30 topics... 9x non-election topics [discarded] 6x political party topics found 15x issue topics found topic weight UKIP Tax & Spend EU NHS ... Task: Identifying election-related topics
Dataset: Social media Over study period 22 Mar – 17 May 86,939 tweets 52,299 users 10,529 hashtags 2,349 domains shared GE-focused tweets, filtered by identifying GE-related hashtags ...of which 15,152 users tweeted one or more URL a proxy for the topics discussed by Twitter users a proxy for the content sources tweeted by Twitter users; e.g., guardian.com
Twitter trends over time Election day Thu 7 May ‘Battle for Number 10’ Two party leaders Q&A Thu 26 Mar ITV Leader’s Debate Thu 2 Apr BBC Opposition Leaders’ Debate Thu 16 Apr BBC Question Time Election Leaders Special Two party leaders Q&A Thu 30 Apr
Temporal correlation between hashtags and topics large circle = significant at 0.01 threshold red = positive correlation (hashtag and topic trend together)
Community 1 Includes: The Daily Mail, The Daily Star, The Express, The Telegraph, The Sun, The Times and Western Morning News Community 4 Includes: Western Mail and Yorkshire Evening Post expressandstar.com order-order.com thunderclap.it express.co.uk thetimes.co.uk dailymail.co.uk electmps-ukip.nationbuilder.com ukip.trendolizer.com conservatives.com countryside-alliance.org leicestermercury.co.uk pensions-insight.co.uk join.ukip.org linkis.com breitbart.com yougov.co.uk westernmorningnews.co.uk cambridge-news.co.uk ukip.org unilad.co.uk nopenothope.blogspot.co.uk yorkshirepost.co.uk blogs.spectator.co.uk telegraph.co.uk thecourier.co.uk thesun.co.uk ukipnw.org.uk labour25.com coventrytelegraph.net itv.com southwalesguardian.co.uk news.channel4.com ebay.co.uk supportchaz.weebly.com plaid.cymru fivethirtyeight.com partyof.wales walesonline.co.uk aberystwythvoice.wordpress.com facebook.com Community 2 Includes: The Daily Mirror and The Independent Community 5 Includes: The Evening Standard labourlist.org channel4.com ipsos-mori.com press.labour.org.uk action.makeseatsmatchvotes.org tompride.wordpress.com bbc.co.uk theconversation.com standard.co.uk news.sky.com thisislocallondon.co.uk cnduk.org politicshome.com thenational.scot vine.co parliament.uk ssl.bbc.co.uk theyvoted.org getbucks.co.uk chroniclelive.co.uk kittysjones.wordpress.com shop.labour.org.uk conservatives.com myukip.com ukipbarnet.org wokinghamukip.org.uk telegraph.co.uk dailymail.co.uk thetimes.co.uk extras.thetimes.co.uk express.co.uk thesun.co.uk dailystar.co.uk Only high-degree nodes are depicted Tax and spend Coalition Benefits Polls Debates EU Media Regions Donations Schools Business Housing Economy London NHS Conservatives SNP UKIP Lib Dems Labour Greens 3 5 7 10 12 1 2 3 4 5 Center is at .6111 news topic proportions (LDA on newspaper news articles)
Summary & Ongoing Work • Social media → communication exposure → network effects • Shared audience communities... ...relate to party ideologies ...relate to newspaper ideologies ...reveal biased issue topics • Missing link: domain to user exposure? Ongoing work.. • GNIP data on GE2015 • MP candidates’ tweets
The next evolution...? 2005 2010 2015 5.5M* 7M UK 15M UK 28M UK UK’s first ‘social media election’ 31M UK The ‘hashtag election’ – Mark Frankel, BBC, A.Ed. of Social News 2020 ? ?
Network bias in social media is different – harder to spot, not controlled by single individual, unpredictable...democratic? Difficult to buy influence when bias and exposure are emergent properties of a complex, interconnected, networked system?
Different news for different views: Political news-sharing communities on social media through the UK General Election in 2015 (at NECO 2016) Thanks for listening! Matthew J. Williams University of Exeter [email protected] http://www.mattjw.net @voxmjw Iulia Cioroianu University of Exeter www.iuliacioroianu.info Hywel T.P. Williams University of Exeter @HywelTPWilliams ExpoNetMeasuring information exposure in dynamic and dependent networks
mattjw
mprove
doradora09
ecoopnet
ishiijunpei
joisino
arumakan
takasumasakazu
genomethica
shimacos
konakalab
dfm
ailaboocu
cromwellryan
orderedlist
skipperchong
sstephenson
maggiecrowley
tammielis
csswizardry
rmw
62gerente
danielanewman
chrislema
malarkey