Post by Farida Vis
As part of the Reading the Riots on Twitter project (part of the larger Reading the Riots project, a collaboration between The Guardian and The London School of Economics), I am part of a team that examined 2.6 million tweets that were donated to the project by Twitter. The larger project seeks to better understand underlying causes of the 2011 UK summer riots. The team is led by Paul Lewis, The Guardian’s Special Projects Editor. During the riots Paul reported on the front-line in cities across England (most notably from Twitter: @paullewis). The Twitter work is led by Professor Rob Procter at the University of Manchester.
We examined this database of tweets focusing specifically on how rumours circulated on Twitter (see this interactive made by The Guardian’s team that visualises the data beautifully, and this write up of the rumour analysis), the function different actors have in propagating and spreading information flows, in addition to establishing whether the platform was used to incite as well as examining other forms of organization. Results from the project were published in The Guardian during six days of extensive coverage in early December 2011 and our social media findings and recommendations have since been widely reported (inter)nationally. For more details of the project, the team and the coverage in The Guardian so far, see the project site.
In this post I want to highlight in more detail some of the methods we used to examine the different types of users who commented on the riots using Twitter. The full set of tweets came from around 700,000 individual users, producing an expected very long tail. In discussion with The Guardian we looked at those accounts that received the most mentions, initially focusing on the top accounts (more than 500 mentions initially, as well as the top 200 and top 1000 most mentioned accounts). I feel that we should also try to look at a range of other samples, to see what is going on in that very long tail or even with those accounts that are frequently mentioned, but do not make the top 1000 accounts (which have at least 182 mentions each). It would be good to be able to compare the top of the tail to the middle and the end as well, with a two more (stratified) randomly selected samples for 500 accounts each for example. My feeling is that doing this will reveal a number of things. As the top accounts mainly contain elite UK users (for obvious reasons), these other samples will add a potentially stronger international dimension (media outlets, journalists, bloggers and so on) as well as more non elite UK based users and offer a more complex picture of how the riots were tweeted.
Top 200 accounts
Returning to the top mentioned accounts, The Guardian Data Store reported on these top 200 accounts as part of the Reading the Riots coverage in December (200 most influential Twitter users during the riots: are you on the list?), causing a brief flurry of activity of people commenting on this list (both in the comment section of the article, but more actively under the main #readingtheriots hashtag. A Topsy archive of nearly 900 tweets using this hashtag and commenting on the project can be found here ). People checked if they had made the list, surprised to be part of it in some cases, and at times expressing how they had fared compared to other, more famous accounts, such as Justin Bieber (number 170), ‘It’s a bit weird I ranked higher than Ed Miliband and Justin Bieber’ (andy87news in the comments to the article).
We were interested in both this list of individual accounts, but moreover how they could usefully be categorised and organised as ‘types’ so that we could make broader readings about the types of accounts and users that were most frequently mentioned and to start exploring what this might mean. To find out the first part of this question, we classified different ‘Actor Types’ based on the top 1000 accounts in the database and came up with twenty types of Twitter users in total.
These mentions could result from all Twitter uses: original tweets, being retweeting, @replies or where the account was simply mentioned as part of a tweet. In developing these different types of accounts, we adopted 12 categories of Twitter user from this recent study led by Gilad Lotan (2011), which had produced a typology of different user types during the Tunisian and Egyptian revolutions.
Taking this typology as a starting point, we adapted it to better suit the specificities of our dataset and (inductively) added eight of our own categories. For example, we wanted to be able to draw distinctions between those users who had a well-established presence on Twitter in the UK – the UK Twitterati (changed from the original ‘digirati’ category) and gauge the role they played in tweeting the riots. We acknowledge that a choice had to be made about which list to use (in this case one produced by The Independent in February 2011, further details here), and that minor differences may exist if another list of influential UK Twitter users had been used instead (suggestions welcome!).
There are a number of mainstream journalists and broadcasters who are highlighted as UK Twitterati on the list from The Independent. Examples in our top 200 accounts include Krishnan Guru-Murthy (@krishgm) and Piers Morgan (@piersmorgan). The list also highlighted journalists from mainly online media for example, Mike Butcher (@mikebutcher), Editor of TechCrunch Europe. His tweets appear frequently in the riots dataset and are widely retweeted as well. The fact that these users are seen as Twitterati however was of importance to us here. As has been already reported elsewhere, the mainstream media dominates this list of top 200 accounts, as is also clear from the chart below.
Another addition to our typology highlights that we wanted to address the not uncommon practice of accounts set up specifically to engage with crisis events. We therefore created a category that highlights those accounts set up directly in relation to the riots (‘riot accounts’, code 12). The importance of one such account, @riotcleanup has already been widely mentioned during the time of original coverage and after.
To summarise, we updated the ‘digirati’ category to ‘UK Twitterati’, added online journalists (highlighting those people working for blogs, news portals or journalistic entities that exist solely online); non-(news) media organisation employees; police/emergency services; riots accounts; members of the public; unclear; account closed and fake/spoof account. Our final list of actors is listed in our Actor Types code frame, including some further details for each category as well as a number of examples per type.
Such ‘Actor Types’ inevitably highlight the difficulties this type of coding can cause on the part of the coder. For example in terms of coding ‘members of the public’ (code 15), we are aware that they are essentially spread across numerous codes, but in keeping these different codes in play, we wanted to highlight the different ways in which Twitter users may present and brand themselves (for example through code 6, non-(news) media organisation employees), highlighting a professional profile. Recent work by Marwick and boyd (2010) has highlighted that due to the public nature of the platform, users actively manage their self presentation (through personal branding), both in how they present themselves in Twitter bios and what they tweet about, imaging an audience in the same way media producers or novel writers might. This means that certain topics are not shared, over sharing information is seen as inappropriate. This imagined audience also varies according to numbers of users people have, those with few users may still speak of an audience of ‘friends’, those with significant followings, ‘micro celebrities’ may be more inclined to talk of communities of followers or indeed ‘fans’. We still know very little about what happens when those perceived audiences change for users or when the circumstances in which they tweet change dramatically, for example when ‘members of the public’ (or indeed journalists in the case of Paul Lewis) are thrust in the social media limelight and follower counts grow exponentially, as we saw with a number of users in our dataset.
To compliment the work of The Guardian Data Store, the bar chart below shows the results of the top 200 accounts, coded in accordance with the code frame described (please click on the image to enlarge it).
In total these top 200 accounts get 567,430 mentioned combined. What is instantly clear from the chart is the overwhelming presence of the mainstream media in this most mentioned list (125768 mentions or 22.2% of 567,430). If we add to this mainstream media journalists (79043 mentions or 14%), only online (news) media (13303 or 2.3%) and online media journalists (4607 or 0.8%), media accounts are mentioned nearly half of the time (44.7%), highlighting the platform as a key source of news in its own right as well as a source of information for the mainstream media to draw on.
Fake or spoof accounts (18163 mentions, 3.2%), set up to offer humorous tweets impersonating somebody else, often a celebrity, are also mentioned as part of this top cited list. The spoof accounts listed in the top 200 included two Harry Potter related accounts: @lord_voldemort7 (number 13, 6697 mentions); @ _snape_ (number 20, 5939 mentions) and the popular @queen_uk account (number 23, 5527 mentions). The presence of such accounts reminds us that Twitter is used by a myriad of different users for wide range of reasons even during crisis moments and that well-established spoof accounts like the ones listed here, continue to be popular during such times. The role of humour was in evidence throughout the dataset, used to highlight a range of different issues (including the status of technology and social media use by various media outlets and government agencies). A frequently retweeted account pretending to run PR for Rupert Murdoch for example tweeted: ‘SkyNews is right. #LondonRiots all Twitter’s fault. Give Twitter to me, we’ll strip it of importance and relevancy like we did to MySpace.’ Comments on other important news issues at the time were frequently linked, most notably the phone hacking scandal and the inability of the police to access BBM (‘just ask News International’). The use of humour and such social commentary as part of wider crisis communication is worth investigating in further detail in the the future.
Very few politicians made this list, either through retweets, @replies, their own active Twitter use, or the mention of their accounts. Those accounts that were listed include the official account for the Prime Minister, the opposition leader, alongside local politicians such as Mike Harris, Labour Councillor for Lewisham Central, which had been a key location for some of the rioting.
The top ten most mentioned accounts were as follows:
- riotcleanup (40960 mentions – coded as riot accounts)
- paullewis (30031 mentions – coded as journalists, mainstream)
- piersmorgan (20412 mentions – coded as UK Twitterati)
- bbcnews (18836 mentions – coded as mainstream media)
- itv_news (15177 mentions – coded as mainstream media)
- bbcbreaking (13476 mentions – coded as mainstream media)
- guardian (11513 mentions – coded as mainstream media)
- lawcol888 (9290 mentions – coded as members of the public)
- simonpegg (9240 mentions – coded as UK Twitterati)
- gmpolice (8904 mentions – coded as police/emergency services)
Although code 12, ‘riot accounts’, is not mentioned as frequently as some of the other actor types discussed above (59193 mentions, 10.4%), the @riotcleanup account is the most single mentioned account in the whole dataset, highlighting the way in which Twitter was used to organise the cleanups in the aftermath of the riots. Moreover, through the high volume of mentions, it shows the reach this account had in a very short amount of time. The person who set up the account Dan Thompson (@artistsmakers, coded as non-(news) media org employees), only narrowly misses the top 10 himself (number 12, with 7033 mentions). Related, and part of the top 10, there is @Lawcol888 (coded as member of the public), with 9290 mentions, whose ‘broom army‘ photo went viral (more info here), again largely due to the further mentions of this image by influential Twitter users like John Prescott (UK Twitterati and actively addressed through an @reply in the original posting of the the image thus bringing it to his immediate attention) as well as the mainstream media picking up on it.
The last account in this top 10, that of the Greater Manchester Police is of specific interest in order to better understand how police and emergency services can use social media, Twittter in particular, during such times. The Greater Manchester Police has a relatively high Twitter profile compared to other UK police forces, largely due to their early and innovative adoption of the platform in order to highlight the work of the force. For example, in October 2010 they used Twitter to report every incident in its area for one day (an archive of these tweets can be found here), which received considerable coverage and interest in their Twitter use.
During the riots a number of tweets specifically encourage Twitter users to follow the Greater Manchester Police account, most notably one by @piersmorgan, who claims it is the best account of the year, as well as other users saying the same. Due to the endorsement from Piers Morgan, included on the UK Twitterati list, this suggestion was potentially seen by nearly one and a half million followers as well as by followers from accounts that retweeted the original endorsement. Aside from the high praise the account received during the riots, in the immediate aftermath of the riots, the same account was criticised for ‘celebrating’ the five-month jail sentence of a woman who had accepted looted clothes (see for this article for example). An apology was issued through their Twitter account. In line with the recommendations we highlight as part of our study, namely for government authorities to make better use of social media during crisis moments, further work on how police and emergency services in particular can make more effective use of social media as well as learn from less successful examples is vital. As Bill Wasik highlights (in this excellent piece on the Riots in Wired magazine), although this is a significant challenge, a sensible approach would look at how offline strategies of ‘community policing’ could be extended to social media, thus building up an online community that will allow such official accounts to disseminate crucial information more effectively during times of trouble.
Being able to analyse not only which information flows were significant and how they circulated, but to also get a clear sense of who propagated the original tweet allows for a further level of analysis that may highlight the presence of certain actors, but not others. Further work is thus needed on the content of the tweets, in order to better understand how some of these accounts were engaged with and indeed how they themselves engaged with other accounts. Although the number of account mentions alone tells us something about a broader picture, to understand this data better, we must also look at the context (context context context!), and the ways in which these accounts have been mentioned. As ‘mentions’ essentially includes all possible Twitter use (from original tweets, to @replies to mentions), this might include a range of different uses and it is important to explore this in more detail. For example some mentions might highlight media accounts that were doing a good job using Twitter in their reporting as well as those that did not. We found examples of both.
Furthermore, accounts that did not make it to the Top 1000 list may still have played a crucial role in a local context, but due to a lower number of mentions across the whole dataset do not show up in a list that is solely based on quantity. For example, the role local journalists play during such events, fewer mentions might still represent significant impact within a local context and this is certainly worth looking into in more detail in the future. While examining the content of the tweets will add to our understanding, speaking to the people behind some these accounts will add further crucial insight still.
Given the extraordinary world events that unfolded during 2011, The Independent has recently produced a new list reflecting this shift, highlighting ‘The most influential non-celebrity users of Twitter during 2011′. This new list highlights many of the accounts included in our riot dataset, both as part of The Independent’s news and international news category. In writing up our data for publication, it is worth reflecting on how the riots put some of these accounts on the map and what this might mean for crisis communication during future events.
In future work it will be important to better understand the role of politicians, the police and other authorities during such crisis moments. Is Twitter always a useful platform to engage with the general public and what are the potential pitfalls for example? Moreover, can better policies for engagement online be formulated so that these can aid communication during times of trouble? If so, what would they look like? These are important issues to understand better. The dataset of riot tweets allows for significant work in this area to be done, and it’s evident that some authorities were clearly better prepared for utilizing Twitter than others.
A recent study by Axel Bruns, Jean Burgess, Kate Crawford and Frances Shaw that examines the use of Twitter (and other social media) during the 2011 South East Queensland Floods in Australia highlights the ways in which the most important emergency service account, the Queensland Police (through their @QPSMedia account) made extremely effective use of the platform. The report highlights that the Queensland Police account played a leading role in disseminating important information to the public quickly, playing a key part in wider discussion. The authors also highlight the ways in which the Twittter account allowed the police to cut through effectively, reaching its audience quickly. Moreover, using the #mythbuster hashtag it played a vital role in directly addressing and correcting rumours and misinformation.
Although the UK riots were not a natural disaster like the floods in Australia, the findings of these two recent studies are certainly complimentary and as the authors of the Queensland Flood study highlight in their recommendation to the wider research community, additional internationally comparative work on social media and crisis communication is necessary, principally in order to share methods, tools, best practice for such ‘big data’ research. I couldn’t agree more. I hope that this current post as well as future ones highlighting insights from the Reading the Riots on Twitter project, can start to make such contributions, not in the least because of the privileged position we find ourselves in with such rich datasets at our disposal.