The Norwegian Twittersphere: Structure and Dynamics

Twitter is by now well established in most developed nations as a key social media platform for public communication. Compared to larger rivals such as Facebook and Instagram, it is seen as especially relevant for relatively fast-paced, close to real-time interactions; this is due especially to its deliberately character-limited message format, which resembles oral communication more than longer-form writing. Additionally, a range of platform design choices – such as the ability to follow the updates posted by other users without, generally, needing those followees’ permission, and the public visibility of almost all tweets, even to visitors to Twitter.com who have not registered an account of their own – have shaped the structure of Twitter’s social network and the nature of user activities on the platform to a considerable degree. While in principle it is possible to set one’s account to ‘private’, so that tweets are visible only to approved followers, only a very small percentage of Twitter users have chosen this option; the remainder are engaging, more or less actively, in a globally distributed public conversation.

This has meant that Twitter has been rightly recognised as highly influential, especially for its role in public information sharing and discussion, in the context of major breaking news, key shared events in society, culture and sports, and the viral distribution of memes (Papacharissi & de Fatima Oliveira 2012; Highfield 2013, 2016; Vis 2013; Hermida 2014), which are often driven and loosely coordinated by one or more central hashtags. In such cases, Twitter users gather together rapidly to form ad hoc (Bruns & Burgess 2015), and sometimes affective (Papacharissi 2014), publics that compile, curate and share information, opinion, memes and other content related to their shared issue of interest; these publics persist for as long as the issue itself is alive, but often dissipate soon after. Indeed, a considerable proportion of Twitter research to date focusses on such temporary publics, and might be categorised under the overall banner of ‘hashtag studies’ (see Rambukkana 2015 for a recent collection of such work).

Such studies have produced valuable insights into the role of Twitter in the context of acute events (Burgess & Crawford 2011), but remain limited in their ability to contextualise their own observations against the broader range of activities taking place on Twitter at any one point. For instance, they might report a substantial absolute number of users participating in the discussion of a breaking news event, but are generally unable to express that number as a percentage relative to the total number of Twitter users active at the time; this means that it is impossible for them to say whether even an event as widely recognised as the Arab Spring (cf. Meraz & Papacharissi 2013) attracted a considerable proportion of Twitter users as followers of or participants in the discussion, or whether the majority of the global Twitter population simply continued their everyday activities without taking much notice of the dramatic events unfolding in Tunisia, Libya, Egypt and elsewhere. Further, the focus of many such studies on a small set of hashtags and keywords in their data gathering also means that they are likely to miss out on relevant tweets that (accidentally or deliberately) fail to include the keywords tracked; this especially includes follow-on @replies that might respond to a hashtagged tweet but do not themselves contain the hashtag again (Burgess & Bruns 2015). Finally, too, the very limited range of immediately available user profile metadata makes it difficult for these studies to target their findings to a set geographical area, even where this would be very valuable – for instance, to examine only those Arab Spring tweets originating in countries of the Middle East and Northern Africa, or to study only those tweets discussing the presidency of Donald Trump that were posted by users based in the United States. This means that the majority of current Twitter research is forced to make only global observations which do not necessarily offer much insight into the uses of Twitter in specific national and local contexts.

This article addresses some of these limitations by taking a new approach to the comprehensive study of an entire national Twittersphere. Its main purpose is to investigate the key characteristics of the historic and present-day Norwegian Twittersphere; in particular, in doing so it also examines the role of distinct national languages such as Norwegian in a globalised social medium where English dominates, and assesses the degree to which ‘echo chambers’ (Sunstein 2009) are affecting networked communication patterns. Norway’s relatively small population of 5 million, its advanced ICT infrastructure and its high usage of online and social media alongside a similarly high usage of public service media and mainstream news media (Syvertsen et al. 2014) position it as a useful case study, whose findings might subsequently be compared with other country-level Twittersphere studies in Europe and beyond. Previous studies of Twitter in Norway and Scandinavia have been concerned with relatively limited groups of accounts and focussed on specific thematic clusters; most typically, they have studied Twitter activities around politicians and journalists (Larsson & Moe 2013; Skogerbø & Moe 2015).

In contrast, by identifying, to the extent that this is possible with the data made available through Twitter’s Application Programming Interface (API), all accounts operated by Norwegian users and institutions, analysing patterns in their public profile information, and mapping their follower/followee connections with each other, we are able to provide new insights into the historical development of the Norwegian Twittersphere, its current network structure and the presence of diverse interests and issues amongst the nearly one million accounts within this community. Our findings are valuable in their own right – and represent only the second national Twittersphere to be mapped in this way, after Australia (Bruns et al. 2014, 2017) – but also constitute important background information for future Twitter research that takes the familiar hashtag studies route: our observations enable such studies to filter their datasets for confirmed Norwegian accounts only, and to examine the presence of accounts with specific interest profiles, as determined by our study, in their datasets.

The major contributions made by this article are twofold. First, and most importantly, we find that the Norwegian Twittersphere is composed of diverse but highly interconnected communities that emerged in earnest in 2009, but received an even more substantial boost in membership when Twitter, Inc. began to offer a Norwegian user interface for the platform in late 2011. Thematic interests in these communities range from business and politics to entertainment and interpersonal communication, with several specialist and minority topics also catered to. The analysis of the overall structure of the Norwegian Twittersphere which we perform in the present article provides a basis for further research into the uses of Twitter in Norway: drawing on our findings, researchers will be able to follow, analyse and compare the activities of selected Norwegian accounts and groups (as selected from the network clusters we have detected, or based on the metrics we have calculated) against other account populations, for example, and thereby to study activity on Twitter for a purposive sample of Norwegian accounts with defined characteristics. This enables research approaches that are considerably different from the still-predominant approach of gathering tweets (and, by extension, accounts) simply because those tweets contain specific keywords or hashtags: instead, they are able to capture all public tweets by these accounts, and thereby to produce a more comprehensive picture of the full range of Twitter activities within a specific community of accounts. We aim to build on the work presented here by pursuing such opportunities, and we encourage other Nordic Twitter scholars to join us in this effort.

Further, the present article documents the methodological steps taken in identifying the accounts that comprise the Norwegian Twittersphere on a large scale, and in analysing their activity and interconnection patterns. Adjusting for variations in national and cultural contexts, we hope that this outline of our methods will enable other research teams to replicate the study presented here for different national or regional Twitterspheres, and perhaps even to apply these approaches to the analysis of other social media platforms.

Our study largely follows the data gathering and filtering process outlined in Bruns et al. (2017). At the time of our data gathering in early 2016, Twitter user IDs were arranged sequentially in order of account creation, counting up from 0 to close to 5,000,000,000. We used the Twitter API to systematically retrieve the profile information associated with each ID, until in February 2016 we found accounts with IDs above 4,900,000,000 that had been created only minutes before we gathered their profile information. This meant that we had exhausted the then current Twitter ID numberspace and reached the leading edge of continuing account creation. It should be noted here, of course, that not all of these IDs are associated with an account: the earliest profile information we were able to retrieve was for Twitter co-founder and CEO Jack Dorsey (@Jack), with ID 12, and we also noted considerable blocks of ID ranges that did not have any users associated with them. In addition to natural churn – where Twitter accounts are created at one point and subsequently suspended or deleted, leaving IDs unassigned – we assume that some such ID ranges were left unassigned for testing purposes at various stages of the development of the Twitter platform. In total, as of late February 2016, our approach gathered profile information for nearly 1.38 billion accounts.

To identify Norwegian accounts within this global userbase, we utilised a number of the available datapoints for account profiles. Contrary to the approach taken by Bruns et al. (2014, 2017), which in part drew on the display timezones users had set for their Twitter profiles, this field proved useless for our purposes: in contrast to the distinct Australian timezones that the previous study was able to utilise, Twitter does not offer a dedicated Norwegian timezone setting, and users are instead forced to use one of the other options representing Central European Time (e.g. ‘(GMT+01:00) Copenhagen’). At the same time, however, we were able to filter our global accounts dataset by the interface language that users had chosen for their accounts (this was impossible for the earlier Australian study, due to the prevalence of English as the primary language in that country).

We further constructed a list of the 100 largest cities in Norway, as well as of the twenty Norwegian counties, and filtered our global data for accounts that mentioned these locations in the free-form ‘Location’ and/or ‘Description’ fields of their profiles. This list was further supplemented with a number of more generic terms, including ‘Norge’, ‘Norsk’, ‘Norway’ and ‘Norwegian’. The first results of this filtering process were manually reviewed in order to identify and eliminate false positives resulting from these filter terms occurring in other contexts; for instance, this was used to distinguish Bergen in the Norwegian county of Hordaland from Bergen op Zoom in the Netherlands, or Oslo, Norway from Oslo, Minnesota. At this stage, we chose to exclude Norwegian cities Alta and Ski from our filter altogether: ‘Alta’ (meaning ‘high’) is a very common term in several Romance languages, and ‘ski’ referred most frequently to the sport rather than the town; in both cases the false positives overwhelmed the true positives. While this could mean that accounts from these locations may be underrepresented in our final dataset, we note that a significant number of such accounts would also have exhibited secondary traits of Norwegianness that we will have identified, including mentions of their counties or of Norway itself (e.g. by using ‘Alta, Finnmark, Norge’ as a location text) or a Norwegian interface language setting.

In total, by using these filters we identified some 621,000 accounts with a language setting of ‘no’ (for the generic Norwegian setting), 10,000 accounts with ‘nb’ (for Bokmål) and four accounts with ‘nn’ (for Nynorsk).

Norwegian has two written standards, Bokmål and Nynorsk.

Further, of course, our dataset constitutes a snapshot of the identifiable Norwegian Twittersphere as of February 2016: many more Norwegian accounts may have been created in the meantime, but equally some of the accounts we did identify at that point may have been deleted or suspended since then, and many other accounts might have been created in the past but already been deleted before we gathered our data and would therefore leave no trace in our dataset. This is an unavoidable aspect of any research that examines as dynamic a space as Twitter or the other major social media platforms.

We begin our analysis of this dataset by examining the historical development of Twitter use in Norway. Based on the account creation dates available for each Twitter account, Figure 1 shows the total number of new Norwegian accounts created per month since the launch of Twitter in March 2006, and still in existence by the time we gathered our dataset in early 2016. The overall pattern of growth in the early years of the platform mirrors similar dynamics at a global level, and as reported in Bruns et al. (2014) for the Australian Twittersphere: Twitter largely remained a niche platform until it emerged to greater media and popular attention in early 2009. We see at this point a sudden increase in new account creations to up to 20,000 new accounts per month (and recognising the possibility of subsequent account deletions, the real number at the time would have been even higher). After this initial spike in sign-ups, the Norwegian Twittersphere continued to grow steadily if slowly for another two years.

Figure 1

Early in 2012, however, Twitter experienced another, even more substantial and rapid influx of new Norwegian users. It is notable here that – unlike the earlier cohort – the vast majority of accounts created from this point onward are today using Twitter with a Norwegian-language Web interface; it is almost certain, therefore, that this new influx is related to Twitter’s launch of a Norwegian interface on 21 December 2011 (@twitter 2011). (The small percentage of older accounts that are now using a Norwegian interface language setting would have switched to Norwegian between December 2011 and our data gathering in 2016.) This pattern demonstrates that even for a country such as Norway, with a population widely able to read and write English, the availability of Twitter in the native language still makes a considerable difference for the platform’s popularity; additionally, of course, it is also likely that such adoption would have been boosted by contemporary Norwegian media reports about the new availability of Twitter in Norwegian. From 2012 onwards, therefore, the total number of new Norwegian accounts created each month rose substantially, and remained at a comparatively high level until 2015; the number of new Norwegian accounts using Twitter in English, by contrast, dropped substantially from its pre-2012 levels.

This pattern was interrupted, however, by a precipitous drop in new account creation between November 2013 and February 2014. Surprisingly, this decline only affected the Norwegian interface language component of the total Norwegian Twitter userbase; new sign-ups of accounts that used Twitter with an English-language interface remained steady throughout this period. We assume, therefore, that the Norwegian interface language option might have disappeared – by accident or by design – from the Twitter Website during this time, and that this served to put off prospective new users who might otherwise have joined during this time. It is notable here that the creation of accounts selecting Bokmål as their interface language commenced around the same time – we speculate, therefore, that the Twitter interface language selection process might have changed to offer dedicated ‘Bokmål’ and ‘Nynorsk’ options, but that – perhaps due to non-Norwegian developers’ confusion over these language choices – the generic ‘Norwegian’ language option was temporarily removed. In turn, the creation of new accounts using a Bokmål interface ceased in July 2015, suggesting another change to the Twitter language options; at the time of writing, the Twitter settings now again offer only a generic ‘Norsk – Norwegian’ option.

Other, smaller fluctuations in new account sign-ups were most likely related to domestic and international events that might have drawn public attention to the utility of Twitter as a platform. Space available in this article does not permit us to examine these in detail; we do note, however, a substantial increase (from below 7,500 new accounts per month to more than 10,000 accounts) in July 2011, and suggest that this indicates a response to the role played by Twitter as a medium for crisis communication during the terrorist attack in Oslo and Utøya that month (Perng et al. 2013; Kaufmann 2015). Such events can have a lasting impact on public attitudes towards social media platforms, especially if the tenor of mainstream media coverage of such platforms also changes as a result.

In a final step in our analysis, we examine the interconnections between the individual clusters we have identified here. This draws especially on the E-I Index, proposed by Krackhardt and Stern (1988), as a measure of the interconnectivity between clusters. First, for each of the clusters we have identified, we are able to count the number of follower/followee connections that members of the cluster direct to other members of the same cluster; we designate these as internal links. Similarly, we are able to count the number of connections directed by members of a cluster to accounts outside of that cluster; those are external links. We note here that all such connections are directional: the links we count here mean that account a follows account b, and by extension that an account in cluster A follows an account in cluster B, but this does not imply that b necessarily also returns the attention and follows a. Aggregated for each cluster, this exercise thus results in a total count of internal as well as external links from accounts in the cluster; these counts are listed in Table 2.

Second, the E-I Index converts these raw counts into a normalised index on a scale from −1 to +1, using the formula

Using this calculation, a cluster whose members exclusively follow other members of the same cluster (that is, whose number of external links is zero) would receive an E-I Index of -1; a cluster with an equal number of external and internal links would receive an E-I Index of 0; and a cluster whose members do not follow each other, and instead only connect to accounts outside of the cluster, would receive +1. Since the very concept of network clusters presupposes a community of nodes that are more densely connected to each other than to the rest of the network, clusters with an E-I Index of +1 are unlikely to occur; nonetheless, the relative differences in E-I Index values between clusters offer a useful measure of how inward- or outward-looking the members of each cluster are.

The analysis of such interconnectivity between clusters is important especially in the context of present debates about ‘echo chambers’ (Sunstein 2009) and ‘filter bubbles’ (Pariser 2012). Unfortunately, the ‘echo chamber’ and ‘filter bubble’ concepts remain poorly defined in current literature: while Sunstein’s books Republic.com 2.0 (2009) and #Republic (2017) have done much to popularise the term, for instance, they do not provide a clear framing of what exactly an ‘echo chamber’ is, or what criteria should be used to detect it. In their work on echo chambers amongst Democrat- and Republican-leaning Twitter users in the United States, Colleoni et al. (2014: 318) suggest that ‘the mechanism through which this fragmentation of political discourse operates is homophily, defined as the tendency of similar individuals to form ties with each other’, and we follow this perspective by seeing echo chambers as related predominantly to the follower/followee connections made by Twitter users; these are most crucial in determining the potential reach of any actual tweets posted or retweeted by these users (in keeping with the metaphor, they circumscribe the distance that the echo can travel without additional amplification).

By contrast, although – in the absence of data on Norwegian accounts’ day-to-day tweeting activities and content – we do not address this concept in detail here, we would then define filter bubbles as related to what those accounts tweet (and in particular, what existing information they share and retweet), as ‘filtering’ implies a conscious case-by-case decision about which incoming information is passed on and which is not; a filter bubble would then emerge when a group of Twitter users (within the same cluster or across multiple clusters) all make the same decisions about what information they pass on to their followers, and what information they ignore. The two concepts are clearly interrelated, of course: what information is available for filtering to a given user depends in part (but not fully) on whom they follow in the network, and how far the information that passes through a user’s filter travels is determined in turn in large part by the structure of their network that follows them. However, in spite of a frequent conflation of the two terms in scholarly and popular literature, we suggest that a useful distinction can be made here: extending the underlying metaphor, the filtering provides (or re-broadcasts) the sound source for the echo, and the echo chamber determines how far that sound may travel.

Proponents of these concepts suggest that the structure and affordances of social media platforms encourage homophilous networking patterns and synchronised filtering practices, and therefore mean that users will increasingly encounter only that subset of all available views and information that is especially likely to agree with their own preexisting interests and preferences – endlessly repeating the same topics, ideas, beliefs, values, and ideologies. At face value, the existence of the thematic network clusters that we have shown for the Norwegian Twittersphere might be seen as supporting this thesis: users who follow accounts from only one of these clusters will see in their Twitter feed only the tweets that circulate within that cluster (but we note that this already assumes that such users do not also use Twitter’s search functions, or follow the hashtags and trending topics that the Twitter platform may also recommend to them). However, if members of the clusters we have identified also follow a substantial number of accounts from outside, the ‘echo chamber’ thesis is undermined: such members would be exposed to a wider range of views and topics than might be available in the cluster itself. Further, if they act on that information by retweeting it to their own followers, this would also challenge the ‘filter bubble’ concept as we have defined it above.

We therefore operationalise the E-I Index as a useful measure of the extent to which such ‘echo chambers’ do exist in observable reality. Table 2 presents both the raw counts of external and internal follower/followee links originating from accounts in each cluster, as well as the E-I Index calculated from these counts using the formula above; as with all other findings presented in this article, these figures cover only links within the Norwegian Twittersphere, and do not include any further follower/followee relationships to non-Norwegian Twitter accounts. It is immediately obvious that, with one exception, the E-I Indices for all clusters are considerably above zero; this shows that, even in spite of the predominant thematic interests expressed in their cluster membership, most users also follow a diverse range of other Norwegian Twitter accounts. The major exception from this pattern is the large ‘Blogs & Humour’ cluster, whose E-I Index remains close to zero: even this indicates only an even balance between inward and outward focus in following patterns, however, and cannot be used as evidence that this cluster would exhibit pronounced ‘echo chamber’ characteristics.

The surprisingly high E-I Index values observed for most of these clusters, however, could be seen as challenging the understanding of these communities of accounts as clusters in the first place: if these groups of accounts are largely following others not in their group, it is reasonable to ask whether they can still be meaningfully understood as belonging to the same group. To address this question, finally, Table 2 also provides an indication of the percentage of accounts in each cluster that are being followed by other members of the same cluster. These percentages indicate that, in all clusters, a substantial subset of all cluster members is followed by their peers in the same community; this is true especially for many of the smaller clusters. For instance, 91 per cent of the 6,055 members of the ‘Photography’ cluster are being followed by other photography enthusiasts, and it is this dense and mutual interconnection amongst members that designates this community as a cluster that can be detected by the Louvain algorithm we have used in our analysis, even if members of this community also follow over 25,000 other Norwegian accounts outside of their own cluster. While these external links are diverse and directed at other clusters across the entire Norwegian Twittersphere, the internal links are considerably more focussed and help the community stand out as a distinct cluster.

Overall, then, the picture that emerges from this analysis is of a truly national Norwegian Twittersphere, without exclusive barriers around specific clusters, in which important information has a reasonable chance to be seen well across the network. This is notably different from past observations of highly polarised network structures, for instance in social media discussions of U.S. politics (e.g. Conover et al. 2011). In addition to the fact that many such studies are built on datasets based on self-selecting hashtag practices, rather than the comprehensive follower/followee network data that we are working with here, this may also be explained by the comparatively much smaller size of the Norwegian Twittersphere, which is likely to work against such polarising tendencies: the fewer communicative partners are available for connection, the less potential there is for cliques to form within the overall network. (In spite of its internal connectivity, the Norwegian Twittersphere as such could still form a kind of national echo chamber if it is poorly connected with the global Twitter network, perhaps; as some two-thirds of the total number of 98.7 million accounts followed by the Norwegian Twitter population are non-Norwegian accounts, however, this does not seem to be the case.)

We note again in this context also that our analysis of the connectivity patterns within the Norwegian Twittersphere extends well beyond the focus on political echo chambers that is common to many other studies. We take instead a broader approach to the ‘echo chamber’ concept, looking for comparatively isolated communities on any topic – and finding few that match this description. As we use the term ‘echo chamber’ in the present discussion, then, we use it to refer to communities of shared topical interests, rather than necessarily of shared sociocultural beliefs or political ideologies. At the same time, we do note that those clusters which address overly political interests are also internally well-connected, with more than 80 per cent of accounts in any of these political or policy clusters being followed by other cluster members: we therefore do not find any strong evidence of internal polarisation within these clusters either. (Such observations could be extended through further analysis of the specific network structures within each of these clusters, which the available space for the present article does not permit us to present here.)

Such percentage metrics, then, also offer an insight into the internal network structure of each cluster. High connectivity percentages within a cluster indicate a comparatively flat internal network structure, where most members have a reasonable chance of being followed by one of their peers; lower percentages point more strongly to a hub-and-spoke structure in which a smaller group of leading accounts have attracted a disproportionate share of their peers’ attention. This is the case especially for ‘Music & Entertainment’, ‘Oslo/National News & Culture’ and ‘Social Elites & Satire’ (the latter of which contains, for example, both official and parody accounts of the Norwegian royal family), where in each case only about one third of cluster members are being followed by their peers. This more centralised structure is not surprising given the themes of these clusters: in each of them, the dominance of a small number of nationally prominent celebrity, institutional and news accounts reflects established patterns of public attention in the Norwegian public sphere well beyond Twitter.

This article has presented the first ever comprehensive analysis of follower/followee network structures in the Norwegian Twittersphere. We have identified a wide range of network clusters that largely represent thematic interests (rather than geographic or demographic distinctions), and outlined the divergent dynamics that have led to the establishment of these clusters and continue to drive their further development. In particular, we have shown how a single key alteration to the Twitter platform itself – the introduction of a Norwegian-language user interface in December 2011 – triggered a substantial influx of new Norwegian accounts to the platform, considerably altering the structure of the network and affecting the balance of personal and professional, private and public interests represented in the Norwegian Twittersphere.

Our overview of the activity metrics for these diverse network clusters has pointed to further significant distinctions: while some clusters in the network are highly active and support the networking and communication between most of their members, others are comparatively quieter and focus attention on a smaller number of highly active, leading accounts, most likely operated by key individuals and institutions. This points to a number of different approaches to using Twitter amongst its Norwegian userbase: while some are using the platform to actively post their own updates and to engage with others, others are there predominantly to follow key information sources and to ‘listen’ to their news (Crawford 2009).

Finally, at least for a broader perspective that focusses on the intersections between overall thematic interests rather than specific sociocultural beliefs or political ideologies, we have also shown that even in spite of current concerns about the supposedly deleterious effects of social media on public debate – expressed in discussions about ‘echo chambers’ and ‘filter bubbles’, and moral panics about ‘fake news’ – there is very little evidence for the existence of such ‘echo chambers’ in the observable structure of follower/followee connections in the Norwegian Twittersphere. Rather, although we can readily detect densely connected regions of the overall network and describe these as communities of accounts that address shared themes and topics of interest, Norwegian accounts generally also choose to follow a large number and diverse range of other participants in the national Twittersphere, avoiding the trap of settling into purely homophilous networks.

We note here, however, that our study is based solely on an analysis of follower/followee connections, and does not take into account the information – the tweets – actually transferred through such connections on a day-to-day basis. We have therefore focussed only on the concept of ‘echo chambers’, which we see as more closely related to such underlying network structures. By contrast, the presence of ‘filter bubbles’ – a concept that we understand as speaking more directly to what users choose to share rather than whom they choose to follow – could be tested by observing (for each of the clusters we have identified, or for a selection of especially relevant clusters) the content of the tweets that accounts post on any given day. Such an observation could examine, for instance, whether different clusters draw their information from different outside sources (by analysing the URLs embedded in tweets) or to what extent cluster members are prepared to retweet content originating from outside of their own cluster. If a given cluster were to draw on a unique set of outside sources, or to retweet only its own members’ tweets, then such a cluster could be said to exhibit ‘filter bubble’ traits. The continuous tracking of the tweets posted by the 988,000 Norwegian Twitter accounts that we have identified here (or at least by a substantial subset thereof) that would be necessary for such an analysis is well beyond the scope of the present article, however, and must for now be left to further research.

The structural analysis of the Norwegian Twittersphere that we have presented here is valuable in its own right – it offers a comprehensive overview of the themes and interests represented in Norwegian Twitter, and may be usefully compared against further such studies that examine other national Twitterspheres, or even against the global Twitter network. It also constitutes a snapshot of the platform’s take-up in Norway at a particular point in time (in early 2016), and a future iteration of the same study should shed useful light on further changes in the Twitter userbase as the platform evolves and as it continues to compete with other social media services. It is just as important, however, to understand the data presented in this article as a crucial background benchmark for further research results: for future studies of Twitter activities in Norway, our observations make it possible not only to provide a general overview of tweeting dynamics (for instance, to indicate that a particular event or topic attracted a given number of unique participants who together generated a set number of tweets), but also to pinpoint which clusters in the network engaged especially strongly with these issues. A political controversy whose discussion does not spread beyond the nearly 17,000 members of the ‘Political Debate’ cluster, for example, might be important, but has failed to attract the interest of anyone but political insiders. By contrast, if that same controversy became the subject of discussion in thematically non-cognate clusters, then it should be understood as more significant: it has connected far more widely and engaged user communities not normally concerned with tracking political developments.

The lack of background data on the underlying structures of specific national Twitterspheres has traditionally made such assessments of the reach of specific issues difficult if not impossible. By comprehensively mapping the structure and historical dynamics of the Norwegian Twittersphere, the research presented here enables such deeper analysis for the first time.