The analysis of content and meta–data has long been the subject of most Twitter studies, however such research only tells part of the story of the development of Twitter as a platform. In this work, we introduce a methodology to determine the growth patterns of individual users of the platform, a technique we refer to as follower accession, and through a number of case studies consider the factors which lead to follower growth, and the identification of non–authentic followers. Finally, we consider what such an approach tells us about the history of the platform itself, and the way in which changes to the new user signup process have impacted upon users.
Determining accession dates
An example: @KRuddMP
Advanced uses: Detecting ‘fake’ followers
While the development of methodological approaches for the study of public communication on Twitter has progressed rapidly in recent years, it has focused primarily on the capture and analysis of tweets, from small–scale samples to very large datasets, up to and including the full ‘firehose’ of all tweets (cf., Bruns and Liang, 2012; Bruns and Stieglitz, 2012; 2013; Gerlitz and Rieder, 2013). Other aspects of the Twitter platform have been considerably less researched, in part also due to the technical difficulties posed by such research: while the tracking of tweets matching keywords and hashtags is comparatively straightforward, a comprehensive analysis of user profiles or follower/followee networks is considerably more difficult (see, e.g., Bruns, et al., forthcoming).
However, the comparative difficulty of examining these other important features of the communicative frameworks provided by Twitter should not lead to their being overlooked. An analysis of public statements — tweets — alone is insufficient without also considering who is there to see them: for example, by examining the follower networks of actively tweeting users to explore the size of their Twitter audience. Indeed, the processes by which accounts develop their audiences of Twitter followers over time are also of interest in their own right, and may lead to the recognition of a number of common Twitter ‘careers’ for different types of users (cf., Horan, 2013).
This paper, therefore, introduces a methodology for the retrospective investigation of Twitter follower accession: the growth of a given Twitter account’s follower base over time. We show how Twitter’s Application Programming Interface (API) may be used to determine not only which users follow a given target account, but also to approximate when they began to follow it, and we draw on a number of leading Australian politicians’ Twitter accounts to demonstrate how this information may be used to identify key events in the Twitter careers of these accounts.
Determining accession dates
The methodology adopted to determine the date at which a user followed our target is, in essence, very simple. A call to the followers/ids function of the Twitter API, specifying a target user, returns that user’s list of followers in descending order; that is, the most recent follower appears at the top of the list. The same behavior can be seen by accessing one’s own list of followers on the Twitter Web site, or indeed in any application which utilizes the Twitter API for this function (e.g., Hootsuite, TweetDeck, Twitter for iOS, etc). Accordingly, for (imaginary) target account @FredBloggs1984, who has 150,000 followers, we know that the account listed at the bottom of the list was the first to follow him (after Mr Bloggs himself joined Twitter — a date contained within his user profile), and the account listed at the top was the most recent follower.
By subsequently calling the ‘users/show’ function of the Twitter API for each ID returned by the initial call, we are able to gather the user profile of each follower, and identify the date on which each follower joined the Twitter platform. Repeating this process for each user who follows @FredBloggs1984, we end up with a list of 150,000 Twitter users, in the order they followed Mr. Bloggs, and the user profile information for each of these followers. For our current purposes, the only information it is necessary to retain from each user profile is the date on which each follower joined Twitter, although the additional metadata for each follower account may also become relevant at a later stage of the analysis.
Following the framework outlined in Hirst (2013), let us assign an accession number a to each of the target account’s followers, counting from 1 (the earliest follower) to the total number of followers (in the example above, 150,000). Let us further designate creationdate(a) as the date on which Twitter account a was created. Given that our followers are numbered in the order in which they began following the target account, we can now state that
- user a followed the target account after creationdate(a), and
- user b followed the target account after user a, for any b > a.
We can therefore also state that
- user b followed the target account after creationdate(a), for any b > a.
For any given user x, we can therefore calculate the earliest possible date at which they could have followed the target account, as the most recent account creation date encountered for accounts 1 through x:
- followdatemin(x) = max(creationdate(1) → creationdate(x))
In essence, this approach means going through the ordered list of followers from follower one onwards, noting down each most recent account creation date we come across, and assigning it as a join date to all later followers until we come across an even more recent account creation date. This approach follows Meeder, et al. (2011), who — defining each follower–target account relationship as an edge — determine the edge creation time “by positing that it is equal to the greatest lower bound that can be deduced from the edge orderings and follower creation times” . They also offer a mathematical proof for the validity of this approach, establishing “record–breakers”: accounts which are significant because they were created after the last known follow date. In our notation above, these are the accounts x for whom
- followdatemin(x) = creationdate(x)
An example: @KRuddMP
A simple visualisation of follower accounts’ accession dates against their creation dates illustrates the underlying principle. Figure 1 presents data collected for the then over 1.2 million Twitter followers of @KRuddMP, the account of former Australian Prime Minister Kevin Rudd, in late June 2013; it plots the accession number a for each follower on the vertical axis, and the corresponding creationdate(a) on the horizontal axis.
It is immediately obvious from this graph that for any range in the total list of followers (that is, for any given value of a), the creation date of a follower account may go back as far as mid–April 2006 (when the Twitter platform itself was created, and the first early adopters signed up); the majority of followers’ accounts were created after early 2009, however, when Twitter achieved a greater popularity both on a global level and especially also in Australia.
Figure 1: Follower accession curve for @KruddMP, in late June 2013.
While for any range of values for the accession number a, the left–hand edge of the graph is thus highly diffuse, its right–hand edge is always clearly defined: it is determined by the ‘record–breakers’ identified by Meeder, et al. (2011), who form the boundary points on the visual representation of the follower accession data. This leading edge of the follower accession graph can be shown in isolation by plotting followdatemin(x) — and its curve represents the growth curve of the target account’s follower base.
Importantly, then, this visualization depends on the fact that the follower list returned by the Twitter API is ordered by accession date; without this ordering, followdatemin(x) could not be calculated. Recent support documents for version 1.1 of the Twitter API state that this list is no longer guaranteed to be in chronological order. To date, however, we have not identified more than a handful of follower accounts per target user which are obviously listed out of order (e.g., followers who are placed to the right of the sharp edge which emerges in graphs such as Figure 1). Given the warning in the API documentation, though, it is possible that the utility of the method we describe here will diminish over time.
On the other hand, as long as the target account and/or the follower accounts themselves remain in continuous use, our approach generates reliable results even if any of these accounts have been renamed at some point: while Twitter accounts may change their names at any point, their follower networks remain the same. We know, for example, that @KevinRuddPM changed its name to @KRuddMP following Rudd’s replacement as Australian Prime Minister in 2010, yet the data shown in Figure 1 extend to a time well before that moment.
There is one significant caveat, however: gathering an account’s list of followers, we only gather information about those followers who at present still remain as followers. If there was a large group of followers who left at some point before we gathered our data, we will not find any evidence of that exodus; we can only see when those followers who have stuck with the target account until now began to follow it. Similarly, if a user followed, unfollowed, and then re–followed the target account, we will only be able to see their presence since the last time they re–followed.
Accordingly, in attempting to predict past histories of Twitter users at particular dates (e.g., the number of followers Julia Gillard had the day before she ousted Kevin Rudd as Australian prime minister), we must allow for some variance between our results and the actual data. That said, however, and as Meeder, et al. (2011) theorise and evaluate, the methodology is only valid in the first place when the user’s follower base is large enough that we are able to predict the accession date (that is, the date a user followed the target account), and as such the margin of error should remain relatively small.
Meeder, et al. (2011) tested the impact of unfollowings on the accuracy of the method outlined here, by randomly deleting followers from a known list, and found that the maximum error which this deliberate reduction in data fidelity introduced measured “no more than six hours” . Although their focus was on celebrity accounts with a substantial number of followers (and thus with very fine–grained accession data), it is clear from such experiments that the methodology survives the disappearance of individual users from an account’s follower list, although of course information specific to those users is lost.
It is also obvious, however, that the fidelity of the follower accession data is inherently connected to the size of the target account’s follower base, since each follower constitutes a data point towards the follower accession curve. Inevitably, for target accounts with fewer followers it will be more difficult to retrace their follower growth with certainty than it is for target accounts with a larger follower base, since only a small percentage of new followers will be situated on the leading edge that determines the shape of the follower accession curve and enables us to estimate the total size of the target account’s follower base at any one point in time. There is no hard and fast rule which requires a target account to have a specific number of followers before our method may be able to be applied, however; rather, it is up to the individual researcher to decide on the amount of approximation and extrapolation they are prepared to accept in each case. Empirically, follower accession curves for accounts with more than 1,000 followers appear to provide a reasonable approximation of the target account’s follower growth.
The practical utility of the method outlined here can be demonstrated by a number of examples which draw on the Twitter accounts of Australia’s political leaders. These examples illustrate factors both internal and external to Twitter which have affected the follower growth of the politicians’ accounts, and thereby point both to the effects of Twitter’s own platform politics (Gillespie, 2010) and to the effects of political events on Twitter usage and users.
The graph in Figure 1 already offers a first point of entry into such analysis. Most notably, it is obvious that Kevin Rudd’s follower growth for the period between June 2009 and January 2010 is significantly greater than that for the remainder of the time covered here. The curve during these months shows a very rapid growth in followers which seems to begin as abruptly as it ends. Between these dates, @KevinRuddPM (as the account was known then, during Rudd’s first Prime Ministership) added more than 700,000 new followers, for no reason that could be identified from the data, or indeed from the Australian political landscape, alone. Such rapid growth may be related to key events in the lives or careers of Twitter users, of course, and sudden increases in follower numbers have been observed and explained in cases, albeit for shorter periods; for example, prominent YouTube contributors @CharlesTrippy and @CoreyVidal both saw significant follower growth after the release of their popular CTFxC Wedding YouTube video (http://www.youtube.com/watch?v=xQVFGdmO5b8), which Vidal produced.
However, in the present case, the time period in question did not correspond to any particular events either in the life of Rudd himself, or in Australian politics. Indeed, Rudd’s follower accession rate was substantially smaller both before and after this time, during periods which included Rudd’s election as Prime Minister and his subsequent loss of the position to a challenge by his deputy, Julia Gillard. This sudden and not immediately explicable increase in Twitter followers even led some of Rudd’s enemies on the hard right of Australia’s political scene to suggest that Rudd had bought ‘fake’ followers in order to appear more popular (see, e.g., Andrews, 2011).
Figure 2 shows Rudd’s estimated follower growth per day, plotting for each day the number of users who share the same followdatemin(x) as we have defined it above:
Figure 2: Follower growth rate for @KevinRuddPM/@KruddMP, per day.
The true explanation for this sharp and sustained rise in the follower growth rate for Rudd’s account is to be found not in the external circumstances of Australian politics, but in the internal policies of the Twitter platform itself. Early in 2009, Twitter introduced the first iteration of its Suggested User List, which featured a handful of major, important, and/or notable Twitter accounts whom new users were all but forced to follow as part of their account sign–up process. Rudd was one of the very few Australians on this (global) list, and — evidently — benefitted significantly. Notably, Meeder, et al. (2011) show similar follower growth effects caused by the Suggested User List during the same period for some of the ‘Twitter celebrities’ whose accounts they use for their evaluation of the method they outline.
In fact, our method allows us to make a reasonably reliable distinction between these newly created followers, who followed Rudd’s account mainly because the Suggested User List was an integral part of the user sign–up process, and other users who began to follow Rudd during the same period. For each user a, we know their account’s creationdate(a), provided by the Twitter API at a second–by–second resolution, and we have extrapolated an estimated followdatemin(a) using the method outlined above. We are therefore able to calculate the difference between these two timestamps, and distinguish two or more subsets of the total follower base using the result of that calculation.
Figure 3 therefore shows Rudd’s daily follower growth rate, but distinguishes the daily growth in those users who followed Rudd within 60 minutes of creating their own account, and in those who took more than 60 minutes after joining Twitter to become followers of Rudd’s account. (The 60 minutes threshold necessarily represents an arbitrary value; we have chosen it to capture even relatively slow sign–up processes while avoiding false positives from users who created their accounts without following Rudd, but then subscribed to his tweets only a short while later.)
Figure 3: Follower growth rate for @KruddMP, per day, distinguishing users who followed Rudd within 60 minutes of account creation from other followers.
The result of this differentiation is striking: between late June 2009 and late January 2010, the influx of users who followed Rudd within an hour of joining Twitter (shown in green) is immense, peaking at over 5,000 followers per day. However, it should also be noted that this number includes only those accounts who still followed Rudd in June 2013, when we gathered the data underlying this Figure 3: we assume that a substantial portion of Twitter users who signed up during this period and followed Rudd because, in effect, they had to choose some first accounts from the Suggested User List quickly unfollowed him again once the account set–up process was completed; Rudd’s actual follower growth rate during this period in 2009/10 may well have been significantly higher still.
At the same time, it is also worth pointing out that even during this period of sharp follower growth, Rudd was also attracting a significant number of Twitter followers who had been active on the platform for more than one hour; as Figure 4 shows, his influx of such more seasoned users remained steady at around 500 new followers per day, roughly commensurate with the periods before and after the sudden influx of newly created follower accounts. (Indeed, as Figure 1 documents, some such new followers had been active since the very early days of Twitter, in 2006.)
In addition to such platform-dependent events, however, it is also possible to trace the impact of external events in the target account’s career on their follower base — at least where they have had a positive impact on follower numbers, since it remains impossible for us to identify users who unfollowed a target account. In order to remove most of the artefacts caused by Twitter’s Suggested User List, we now remove any users who followed Rudd within 90 minutes of joining Twitter; by reviewing the daily growth of Rudd’s remaining follower base over the whole period covered by our data, it now becomes possible to observe a range of key points within Rudd’s Twitter career and to relate them to political events, similar to the YouTube events described above.
Figure 4: Follower growth rate for @KruddMP, per day, showing only users who followed Rudd at least 90 minutes after joining Twitter.
Although the 90–minute filter applied to the follower data does not fully remove the artefacts caused by Twitter’s Suggested User List, we can now more clearly identify several other notable spikes in Rudd’s follower growth rate. These are closely aligned with various key events in recent Australian politics, including the 2010, 2012, and 2013 leadership spills involving Rudd and his successor as Prime Minister, Julia Gillard (in the case of 2012, Rudd’s resignation as Foreign Minister, which preceded it, also emerges as a key moment). There is also a notable lull in follower accession following the 2010 election which returned Julia Gillard as Prime Minister, indicating perhaps a momentary ‘honeymoon’ phase for Gillard and a corresponding lack of interest in Rudd’s return as PM.
Each of the major events identified here leads to Rudd’s follower growth rate accelerating to some 1,000 to 2,000 followers per day, from a comparatively steady average of some 300 followers/day; this suggests both that Twitter users respond to major breaking news by following the accounts of the persons at the centre of the news reports, and that a substantial number of such users continue to follow these persons even once the immediate breaking news moment is over (otherwise these new followers would no longer appear in our data). Such a follower influx need not always be triggered by acute crisis events, however: we can also identify a sharp spike on the day that Australian media reported that @KRuddMP had surpassed the one million followers mark; this suggests that followers beget followers. Repeated over a broader range of leading accounts, as we have done in Bruns (2013), the correlation of such patterns offers useful new insights into the resonance of political and other events on Twitter.
Advanced uses: Detecting ‘fake’ followers
Not all follower growth is genuine and legitimate, however. Indeed, in our original attempt to understand Rudd’s sudden follower influx in 2009/10, we first explored the popular theory, noted above, that the majority of the 700,000 new followers added during that period were in fact ‘fake’ followers: either Twitter accounts created to boost Rudd’s own numbers, or accounts of Twitter spammers who followed Rudd in an attempt to appear more legitimate and reduce the chance of being identified and deleted by Twitter’s anti–spam measures.
However, our subsequent examination of Rudd’s follower growth patterns, outlined above and corresponding to Meeder, et al.’s (2011) observation of their target accounts’ growth patterns, provides the much more convincing explanation that the effects of Twitter’s Suggested User List are responsible for the follower influx observed for @KRuddMP. In other cases, though, follower growth patterns are not as easily explicable without suggesting deliberate interference by interested parties. In mid–August 2013, for example, reports began to appear in the Australian media of a sudden growth in then–Opposition Leader Tony Abbott’s Twitter follower numbers, with some 60,000 new followers for @TonyAbbottMHR emerging around 11 August 2013.
Contrary to the observations made for Rudd’s account, our analysis quickly confirms these to be ‘fake’ accounts whose following behavior is fundamentally out of step with the patterns observed in any other case. Figure 5 shows Abbott’s follower accession graph, analogous to that for Rudd shown in Figure 1:
Figure 5: Follower accession curve for @TonyAbbottMHR, on 12 August 2013, distinguishing ‘fake’ from ‘genuine’ followers.
For the most part, Abbott’s follower accession curve appears ‘normal’: a slow start with several periods of comparative follower growth acceleration that link to events in Abbott’s political career. The relatively strong growth in the most recent weeks covered by Figure 5 is logically related to his role as the conservative candidate in the coming Australian federal election, increasing the mainstream and social media spotlight on Abbott and bringing new followers of various account ages to his Twitter account. However, the most recent growth period, highlighted in red in Figure 5, is very different.
For this period, the follower accession curve is virtually vertical, indicating an extremely rapid growth in new followers — and those new follower accounts were created almost without exception within the last 60 days before we gathered our data on 12 August 2013. Indeed, the account creation dates for Abbott’s most recent followers are so systematic that there appear to be two banks of new followers, one at around 60 days of age, the other about half as old.
This is clearly suspicious, and a significantly different type of behavior to the follow–during–signup process driven by Twitter’s Suggested User List, as discussed above. Abbott’s newest followers did not follow his account when they joined Twitter; instead, they waited almost exactly 30 or 60 days to do so. This suggests a significant element of pre–planning: there is no logical, benign explanation for some 60,000 Twitter users, all of whom joined the platform at almost exactly the same points in time, to follow @TonyAbbottMHR in unison.
The only realistic scenarios to explain these patterns appear to be that a) someone from the pro–Abbott camp decided to boost the Opposition Leader’s numbers on Twitter by buying a number of fake followers for his account; b) one of his opponents did the same in order to exploit the media coverage which resulted from this suspicious influx as an opportunity to embarrass Abbott; or, c) independent of any political motives, a fake follower network operator decided to give their zombie accounts some appearance of legitimacy by connecting them with a genuine, prominent account. (Immediately following the media coverage of his suspicious boost in followers, Abbott’s office worked with Twitter Australia to have these ‘fake’ followers removed from his account.)
Building on the work of Hirst (2013) and Meeder, et al. (2011), this paper has outlined a method to a) acquire follower accession data; b) predict the date at which a particular user followed a particular account; c) interpret follower accession charts as a reflection of key events internal and external to Twitter; and, d) utilize this methodology in order to identify suspicious Twitter followers and their attempts to interfere with follower numbers.
This methodology contributes significantly to the existing body of methods for Twitter analytics and analysis, which have — until now — largely concentrated on understanding how Twitter users congregate around a particular topic, interact amongst each other through @mentions and retweets, or more broadly use the platform to create and view content. However, it goes without saying that the audience and market for that content would not exist without each Twitter account’s circle of followers, and so the present method, and ongoing work to extend it, are especially significant for understanding the drivers of audience development on Twitter (highlighting particularly the differences between internally and externally motivated audience growth), as well as for beginning to differentiate between genuine and non–genuine audience members.
We have outlined a number of caveats to this methodology, particularly around the need to have sufficient data to predict accession dates, and further work is required to establish a minimum baseline at which the present approach is reliable. The method outlined here is not an end in itself. Our approach requires substantial research beyond the follower data as such in order to explain the accession growth patterns it enables us to observe. This is one area where a combination of qualitative and quantitative approaches is crucial in order to establish the cause for particular periods of growth (or otherwise).
The most substantial limitation, and one which it may not be possible to overcome for historic events, is the lack of information available from Twitter on users who have unfollowed the target account, or who unfollowed and then refollowed in the past. Indeed, even to track such unfollow events in current data would require the follower data for any target account to be captured on a regular basis in order to identify any earlier followers who are no longer present in more recent follower lists, with the precision of our observation of unfollow events directly proportionate to the rate at which the follower data can be collected. For large accounts with millions of followers, this is a time– and bandwidth–intensive process, and one which it is not feasible to repeat regularly.
Finally, one significant case where the margin of error from mass unfollowings may be so substantial as to problematize conclusions exists in the case of accounts which see strong seasonal growth and decline; these include the accounts for major television or sporting events (for example, the accounts of @CBSBigBrother and @Wimbledon may demonstrate this behavior). However, it remains to be established whether these accounts are actively unfollowed when they are no longer relevant, or whether follower relationships with these accounts merely lie dormant until they are again tweeting information that their follower base wants to read.
About the authors
Dr. Axel Bruns leads the QUT Social Media Research Group. He is an Associate Professor in the Creative Industries Faculty at Queensland University of Technology in Brisbane, Australia. Bruns is the author of Blogs, Wikipedia, Second Life and beyond: From production to produsage (New York: Peter Lang, 2008) and Gatewatching: Collaborative online news production (New York: Peter Lang, 2005), and a co–editor of A companion to new media dynamics (Chichester: Wiley, 2013) and Uses of blogs (New York: Peter Lang, 2006). He is a Chief Investigator in the ARC Centre of Excellence for Creative Industries and Innovation. His research Web site is at snurb.info, and he tweets as @snurb_dot_info.
Darryl Woodford is a Research Fellow in the ARC Centre of Excellence for Creative Industries & Innovation (CCI), Queensland University of Technology. He has a background in engineering and game studies, including research on the agency of avatars in virtual environments. His current research includes work on social norms and regulation in the video game and gambling industries, and he is leading the development of new digital methods for measuring and evaluating television audience engagement using social media analytics.
Troy Sadkowsky is Big Data developer at ARC Centre of Excellence for Creative Industries & Innovation (CCI) at Queensland University of Technology.
1. Meeder, et al., 2011, p. 518.
2. Meeder, et al., 2011, p. 520.
Tim Andrews, 2011. “EXCLUSIVE INVESTIGATION: Kevin Rudd buying Twitter followers to boost leadership bid?” Menzies House (12 September), at http://www.menzieshouse.com.au/?p=2676, accessed 20 November 2013.
Axel Bruns, 2013. “Follower accession: How Australian politicians gained their Twitter followers,” Mapping Online Publics (8 July), at http://mappingonlinepublics.net/2013/07/08/follower-accession-how-australian-politicians-gained-their-twitter-followers/, accessed 20 November 2013.
Axel Bruns and Stefan Stieglitz, 2013. “Towards more systematic Twitter analysis: Metrics for tweeting activities,” International Journal of Social Research Methodology, volume 16, number 2, pp. 91–108.
doi: http://dx.doi.org/10.1080/13645579.2012.756095, accessed 20 November 2013.
Axel Bruns and Stefan Stieglitz, 2012. “Quantitative approaches to comparing communication patterns on Twitter,” Journal of Technology in Human Services, volume 30, numbers 3–4, pp. 160–185.
doi: http://dx.doi.org/10.1080/15228835.2012.744249, accessed 20 November 2013.
Axel Bruns and Eugene Yuxian Liang, 2012. “Tools and methods for capturing Twitter data during natural disasters,” First Monday, volume 17, number 4, at http://firstmonday.org/article/view/3937/3193, accessed 20 November 2013.
doi: http://dx.doi.org/10.5210/fm.v17i4.3937, accessed 21 March 2014.
Axel Bruns, Jean Burgess, and Tim Highfield, forthcoming. “A ‘big data’ approach to mapping the Australian Twittersphere,” In: Katherine Bode and Paul Arthur (editors). (Re)purposing the (digital) humanities: Research, methods, theories. Basingstoke: Palgrave Macmillan.
Carolin Gerlitz and Bernhard Rieder, 2013. “Mining one percent of Twitter: Collections, baselines, sampling,” M/C Journal, volume 16, number 2, at http://journal.media-culture.org.au/index.php/mcjournal/article/view/620, accessed 20 November 2013.
Tarleton Gillespie, 2010. “The politics of ‘platforms’,” New Media & Society, volume 12, number 3, pp. 347–364.
doi: http://dx.doi.org/10.1177/1461444809342738, accessed 21 March 2014.
Tony Hirst, 2013. “Estimated follower accession charts for Twitter,” OUseful (5 April), at http://blog.ouseful.info/2013/04/05/estimated-follower-accession-charts-for-twitter/, accessed 20 November 2013.
Tyler J. Horan, 2013. “‘Soft’ versus ‘hard’ news on microblogging networks: Semantic analysis of Twitter produsage,” Information, Communication & Society, volume 16, number 1, pp. 43–60.
doi: http://dx.doi.org/10.1080/1369118X.2011.649774, accessed 21 March 2014.
Brendan Meeder, Brian Karrer, Amin Sayedi, R. Ravi, Christian Borgs, and Jennifer Chayes, 2011. “We know who you followed last summer: Inferring social link creation times in Twitter,” WWW ’11: Proceedings of the 20th International Conference on World Wide Web, pp. 517–526.
doi: http://dx.doi.org/10.1145/1963405.1963479, accessed 20 November 2013.
Received 2 February 2014; accepted 5 March 2014.
Copyright © 2014, First Monday.
Copyright © 2014, Axel Bruns, Darryl Woodford, and Troy Sadkowsky.
Towards a methodology for examining Twitter follower accession
by Axel Bruns, Darryl Woodford, and Troy Sadkowsky.
First Monday, Volume 19, Number 4 - 7 April 2014