Mapping Online Publics: New Methods for Twitter Research
Mapping Online Publics: New
Methods for Twitter Research
Axel Bruns, Jean Burgess, and Darryl Woodford
ARC Centre of Excellence for Creative Industries and Innovation
Queensland University of Technology
a.bruns / je.burgess / dp.woodford @ qut.edu.au
SOCIAL MEDIA RESEARCH AND ‘BIG DATA’
• Social media as the ‘big data’ moment in HASS
• But ‘big data’ + ‘social media’ almost always =
• Computational social science – e.g. MSR NYC;
epidemiology; election & stock market prediction
Scott A. Golder*,
Michael W. Macy (2011)
Diurnal and Seasonal
Mood Vary with Work,
Sleep, and Day length
Across Diverse Cultures,
Science 333 (6051):
SOCIAL MEDIA RESEARCH AND ‘BIG DATA’
• ‘Computational turn’ in new humanities research: shift
from computational tools to a new computational
paradigm (Berry, 2012).
• Eg shift from ‘close’ to ‘distant’ reading (Moretti);
‘software studies’ (eg Fuller, 2008) and ANT
approaches to new media platforms; ‘natively’ digital
methods to diagnose patterns of social change
• Intersections between data-driven social media
research & critical platform studies (Gillespie)
THE MAPPING ONLINE PUBLICS PROJECT
• Australian Research Council (ARC) Discovery Project (2010-13) – $410,000 +
ARC Centre of Excellence for Creative Industries and Innovation (CCI)
– First comprehensive study of Australian social media use
– Computer-assisted cultural analysis: tracking, mapping, analysing networked publics
– media, communication & cultural studies’ concerns; combining quantitative analytics
with qualitative analysis
• Mapping public engagement around politics, crisis, culture on Twitter as part of
the broader media ecology
• Various spin-offs: ARC Linkage, LIEF, Future Fellowship, ATN-DAAD, etc.
HASHTAG PUBLICS: #EGYPT @MENTIONS
1-28 Feb. 2011 15 June to 15 Sep. 2011
@mentions between users tweeting predominantly in Latin (blue) vs. Arabic (green) characters
• How do personal publics (cf. Jan Schmidt) around social media
accounts form and dissolve?
– No clear data available through Twitter API
– But: API delivers lists of followers/followees
– And: account creation date for such accounts is known
Method for approximating follower accession
(cf. Bruns, Woodford, Sadkowsky, First Monday 19.4)
• Shows impact of key events on follower growth
• Shows following strategies of central account
• Can indicate influx of ‘fake’ followers (followback bots, etc.)
• Offers comparative tracking across selected population of accounts
• Unable to determine un/refollowing
TWITTER AND TV: TELEMETRICS
• The Importance of Hype
• Social Media Audiences
• From Sabermetric to Telemetrics
• The HypometerTM
• Work with Katie Prowd and partially funded by
THE IMPORTANCE OF HYPE
• Traditional ratings measure
what people have watched, but
have limited impact on what
people *will* watch.
• That is, they are divorced from
the “decision moment”.
• Yet, companies spend millions
promoting shows and
attempting to influence viewer
behaviour, both through TV ads
and through social media.
• How do we measure that?
SOCIAL MEDIA RATINGS
• Contemporary commercial social media ratings have the same limitation,
even if you trust their numbers
BIG BROTHER USA VS AU (AUDIENCE)
• In US, for Big Brother (&
shows generally), there is little
correlation between viewers
• In Australia, for Big Brother
and other shows in our pilots,
high correlation between
viewer count and tweets.
• Applying US models to
Australia is not possible.
• Algorithms need to adjust for
their local environment and
SEASONAL MODELS • Blue Line represents the
ratio of total viewers,
Orange Line represents ratio
of tweets (to season average
• In both one run seasons
(top) and those with mid-
season break (bottom),
tweets are highly
exaggerated version of
traditional ratings model.
• In other words: Users tweet
much more around
premieres & finales than
regular shows. Metrics must
account for this.
• iOS app developed as functional prototype
to act as a ‘modern TV Guide’ for Australian
• Calculates ‘hype’ via a proprietary algorithm
which accounts for national and industry
• Ongoing evaluation of both hype figures and
predictions vs. post-show TV ratings and
social media engagement.
• Clear trend towards ‘dynamic’ audiences; a
proportion of the population on whom
broadcasters should focus.
TWITTER AND THE NEWS: ATNIX
• Australian Twitter News Index (ATNIX):
– Long-term project to track all tweets sharing links to Australian news sites
– Tracking, link extraction, processing, analysis
• Outcomes include:
– Day-to-day (second-by-second) volume of links being shared
– Course-of-day / course-of-week activity patterns
– Trending stories, long-term issues
– Distribution of attention across sites, and change over time
– Impact of paywalls and other changes to site structure and functionality
– Mindshare? Marketshare?
• Further extensions:
– Translation to other mediaspheres: DETNIX, NOTNIX, SWETNIX
– Comparison with other platforms/practices: Facebook, Hitwise, internal data
TWITTER IN AUSTRALIA (AND BEYOND)
• Putting Twitter into context:
– Do specific activities reach beyond pre-existing networks?
– Can we benchmark the phenomena we observe?
– How do individual events compare to each other?
• Research needs:
– Background information on underlying follower/followee network
structures in Australia
How (and how far) does information travel, whom does it reach?
– Baseline data on everyday average Twitter activity patterns in Australia
How extraordinary are extraordinary events, compared to “normal”?
– Standardised metrics for a wide range of events and occurrences
Do similar events unfold similarly? Can we use this to identify them?
Marketing / PR
~120,000 Australian Twitter users
(of ~950,000 known accounts by early 2012)
colour = outdegree, size = indegree
• Data access, platform volatility
• Research ethics
– textual research using public texts, or ‘human subjects’ research
using personal data?
– Consequences of public/personal convergence, data
markets/open data movement, and ‘context collapse’ (boyd)
– Cross-national and cross-disciplinary differences, need for public
• Better integration with existing social and cultural theory
& empirical work
– Mixed methods, especially integration of qualitative and
• Propagation and regularisation of methods (and
consequences for research training)
– ‘Code literacy’ sufficient to engage with the material
consequences of software platforms