1 Introduction
#ArsonEmergency
. Using an exploratory mixed-method approach applied to temporal phases of the discussion, our analyses identified differences in behaviour and content between the Supporters of the ‘arson narrative’ and the Opposers who countered it with fact check articles and official information. This analysis also included the context of the broader discussion by participants Unaffiliated with either polarised group. The first of three phases ended with the publication of a ZDNet article reporting preliminary research on the discussion, which revealed anomalous levels of bot activity (Stilgherrian 2020).1.1 The ‘Black Summer’ bushfires and misinformation on Twitter
-
The bushfires were a topic of much discussion on Twitter, which influenced media coverage.
-
The Supporter and Opposer communities differed significantly in their interpretation of the ongoing events.
-
False narratives and misinformation present in the discussion, which we label the ‘arson narrative’, was promoted primarily by Supporters, and included that:
#ArsonEmergency
and replicate findings reported in ZDNet (Stilgherrian 2020). In the process, analyses revealed two clearly polarised groups in the retweet network, distinctly different in behaviour and content. Supporters promoted the narrative that Australia’s ‘Black Summer’ bushfires were primarily caused by arson, relying on misinformation and biased reporting, while the Opposers countered with official announcements and fact-check articles. We determined that the Opposers responded to the growing Supporter and Unaffiliated account activity on #ArsonEmergency
, when it was exposed in a ZDNet article (Stilgherrian 2020). The publication of the ZDNet article was the trigger that drew the attention of the mainstream media, which promulgated the exposé further, drawing in many more Unaffiliated accounts and noticeably changed the nature of the discussion. Using a different bot detection system, many fewer bots were found than in the analysis reported in the ZDNet article.1.2 Related work
#ArsonEmergency
were repeated in the US during Californian wildfires in mid-2020, even causing armed vigilante gangs to form to counter non-existent Antifa activists who were blamed for the fires on social media.5 Arson was again blamed for the 2021 fires around the Mediterranean, throughout southern Europe and in northern Africa,6 even as the United Nations’ Intergovernmental Panel on Climate Change released its sixth Assessment Report stating that humans’ effect on climate is now ‘unequivocal’ (IPCC In Press). Furthermore, when the misinformation involved relates to conspiracy theories involving public health measures during a global pandemic, the risk is that adherents will turn away from other evidence-based policies, as we see with vaccine hesitancy (Ball and Maxmen 2020), adoption of flat earth beliefs (Brazil 2020) and other conspiratorial anti-government sentiments (The Soufan Center 2021).#ArsonEmergency
was, in fact, created deliberately ( Graham and Keller 2020, argue this), thereby establishing a data void into which disinformation based on the arson narrative could be allowed to flourish before linking it to broader discussions.1.3 Research questions
#ArsonEmergency
discussion:2 The data and its timeline
#Brexit
.Dataset | Tweets | Accounts | Collection method |
---|---|---|---|
Primary | |||
ArsonEmergency | 27,546 | 12,872 | Twarca searches on 8, 12, and 17 January |
Comparison | |||
AustraliaFire | 111,966 | 96,502 | Twarc searches on 8 and 17 January |
#Brexit | 187,792 | 78,216 | Streamed with RAPID (Lim et al. 2018) |
2.1 The timeline
-
Phase 1: Before 6am GMT, 7 January 2020;
-
Phase 2: From 6am to 7pm GMT, 7 January 2020; and
-
Phase 3: After 7pm GMT, 7 January 2020.
#ArsonEmergency
hashtag; the publication of a Conversation article clarifying the ZDNet findings (Graham and Keller 2020); and the clear subsequent diurnal cycle.
#AustraliaFires
, #ClimateEmergency
, #bushfires
and #AustraliaIsBurning
.2.2 Growth of the discussions
#Brexit
discussion lacks a clear intervention event and so its growth is smooth and consistent.9 In contrast, ‘AustraliaFire’ discussion appears to be a hashtag campaign instigated by people in Pakistan and Germany resulting in 45k retweets. Many of the retweeting accounts were suspended, so it is possible they were driven by botnets, and the campaign stops growing suddenly after a few days. The ‘ArsonEmergency’ dataset’s growth pattern clearly shows the point of the intervention, but it continues to grow for several more days after the initial response.
3 Polarised communities
3.1 Community timelines
#ArsonEmergency
was steadily accruing Supporters until the ZDNet article (Stilgherrian 2020), at which point the community was established and remained active for several days into Phase 3. The Opposer community joined almost entirely in Phase 2, and its activity was mostly confined to that phase, while the Unaffiliated continued to join the discussion well into Phase 3. The publication of the ZDNet article appears to have drawn in large numbers of Opposers and Unaffiliated, while the Supporter growth immediately plateaued.
3.2 Behaviour
Group | Tweets | Accounts | Hashtags | Mentions | Quotes | Replies | Retweets | URLs |
---|---|---|---|---|---|---|---|---|
Phase 1 | ||||||||
Supporters | ||||||||
Raw count | 1573 | 360 | 2257 | 1020 | 185 | 356 | 938 | 405 |
Per account | 4.37 | – | 6.27 | 2.83 | 0.51 | 0.99 | 2.61 | 1.13 |
Per tweet | – | – | 1.43 | 0.65 | 0.12 | 0.23 | 0.60 | 0.26 |
Opposers | ||||||||
Raw count | 33 | 21 | 100 | 5 | 8 | 2 | 20 | 9 |
Per account | 1.57 | – | 4.76 | 0.24 | 0.38 | 0.10 | 0.95 | 0.43 |
Per tweet | – | – | 3.03 | 0.15 | 0.24 | 0.06 | 0.61 | 0.27 |
Phase 2 | ||||||||
Supporters | ||||||||
Raw count | 121 | 77 | 226 | 64 | 11 | 29 | 74 | 24 |
Per account | 1.57 | – | 2.94 | 0.83 | 0.14 | 0.38 | 0.96 | 0.31 |
Per tweet | – | – | 1.87 | 0.53 | 0.09 | 0.24 | 0.61 | 0.20 |
Opposers | ||||||||
Raw count | 327 | 172 | 266 | 34 | 7 | 14 | 288 | 31 |
Per account | 1.90 | – | 1.55 | 0.20 | 0.04 | 0.08 | 1.67 | 0.18 |
Per tweet | – | – | 0.81 | 0.10 | 0.02 | 0.04 | 0.88 | 0.09 |
Phase 3 | ||||||||
Supporters | ||||||||
Raw count | 5278 | 474 | 7414 | 2685 | 593 | 1159 | 3212 | 936 |
Per account | 11.14 | – | 15.64 | 5.66 | 1.25 | 2.45 | 6.78 | 1.97 |
Per tweet | – | – | 1.40 | 0.51 | 0.11 | 0.22 | 0.61 | 0.18 |
Opposers | ||||||||
Raw count | 3227 | 585 | 3997 | 243 | 124 | 95 | 2876 | 359 |
Per account | 5.52 | – | 6.83 | 0.42 | 0.21 | 0.16 | 4.92 | 0.61 |
Per tweet | – | – | 1.24 | 0.08 | 0.04 | 0.03 | 0.89 | 0.11 |
Overall | ||||||||
Supporters | ||||||||
Raw count | 6972 | 497 | 9897 | 3769 | 789 | 1544 | 4224 | 1365 |
Per account | 14.03 | – | 19.91 | 7.58 | 1.59 | 3.11 | 8.50 | 2.75 |
Per tweet | – | – | 1.42 | 0.54 | 0.11 | 0.22 | 0.61 | 0.20 |
Opposers | ||||||||
Raw count | 3587 | 593 | 4363 | 282 | 139 | 111 | 3184 | 399 |
Per account | 6.05 | – | 7.36 | 0.48 | 0.23 | 0.19 | 5.37 | 0.67 |
Per tweet | – | – | 1.22 | 0.08 | 0.04 | 0.03 | 0.89 | 0.11 |
Unaffiliated | ||||||||
Raw count | 16,987 | 11,782 | 22,192 | 3474 | 615 | 1377 | 14,119 | 1790 |
Per account | 1.44 | – | 1.88 | 0.29 | 0.05 | 0.12 | 1.20 | 0.15 |
Per tweet | – | – | 1.31 | 0.20 | 0.04 | 0.08 | 0.83 | 0.11 |
3.2.1 Interaction networks
Network | Group | Nodes | Centrality | |||
---|---|---|---|---|---|---|
Betweenness | Closeness | Degree | Eigenvector | |||
Replies | Supporters | 231 (14.6%) | 0.000181 | 0.002871 | 0.004551 | 0.001307 |
Opposers | 82 (5.2%) | 0.000019 | 0.003453 | 0.002757 | 0.001811 | |
Mentions | Supporters | 284 (9.6%) | 0.000304 | 0.005525 | 0.004207 | 0.006575 |
Opposers | 140 (4.7%) | 0.000018 | 0.005067 | 0.001997 | 0.006625 | |
Quotes | Supporters | 169 (18.5%) | 0.000012 | 0.001876 | 0.006170 | 0.016033 |
Opposers | 80 (8.7%) | 0.000005 | 0.003334 | 0.004171 | 0.007302 |
Network | Polarised groups only | Broader network | |||||||
---|---|---|---|---|---|---|---|---|---|
Nodes | Edges | E-I Index | Nodes | Edges | E-I Index | ||||
Supporters | Opposers | Total | Supporters | Opposers | All | Total | Supporters | Opposers | |
Retweet | 493 | 592 | 6645 | − 0.98731 | − 0.99139 | 12,076 | 21,526 | − 0.70961 | − 0.88997 |
Reply | 247 | 105 | 476 | − 0.33333 | − 0.50000 | 2041 | 3031 | 0.62030 | 0.40541 |
Mention | 288 | 149 | 968 | − 0.24615 | − 0.03448 | 3206 | 7523 | 0.69557 | 0.78723 |
Quote | 190 | 104 | 330 | − 0.61832 | − 0.82353 | 1268 | 1542 | 0.45501 | 0.10791 |
3.2.2 The concentration of voices
3.3 Content dissemination
3.3.1 Hashtags
#AustraliaFires
, #ClimateEmergency
and a prominent media owner. These findings have now been confirmed statistically. Even though Supporters used approximately the same number of hashtags per tweet as Opposers (2.92 compared with 2.89), they used 40.9 hashtags per account, including 1.30 unique hashtags per account. In contrast, Opposers only used 17.5 hashtags per account, including 0.36 unique ones. This indicates the pool of hashtags used by the Opposers was much smaller than that of Supporters. The distribution of hashtag uses for the ten most frequently used by each group (which overlap but are not identical), omitting the ever-present #ArsonEmergency
, is shown in Fig. 10. It indicates that Opposers focused slightly more strongly on a small set of hashtags, while Supporters spread their use of hashtags over a broader range (and thus their use of even their most frequently used hashtags is less than for Opposers). Unaffiliated accounts used their frequently used hashtags more often than both groups by the 4th hashtag, possibly due to the much greater number of accounts being active but less focused in their hashtag use. A second hashtag appeared in fewer than 20% of each groups’ tweets.
3.3.2 External URLs
twitter.com
) as arson NARRATIVE aligned, CONSPIRACY content, DEBUNKING content and OTHER. The categorisation was based on the perceived intent of the use of the article, rather than purely on their factual content. Examining the frequency of the ten most shared URLs by each group in each phase, we found Unaffiliated accounts shared mostly NARRATIVE URLs in the first phase, but mostly DEBUNKING articles in the final phase (p.11, Weber et al. 2020).3.4 Coordinated dissemination
3.4.1 Co-retweet analysis
3.4.2 Co-hashtag analysis
#ArsonEmergency
from our co-hashtag analysis. The two largest components discovered highlight the polarisation between the Supporter and Opposer communities (Fig. 14). The ring formation amongst the Supporters and small node sizes indicate less activity including a wider variety of hashtags. Opposers are more active and focused in the hashtags they used. These findings emphasise the findings in Sect. 3.3.1 but also highlight the support of Unaffiliated accounts, the most active of which appear to support the Opposers.
3.4.3 Co-URL and co-domain analysis
bit.ly
. One bit.ly
link appeared much more frequently than others, and it resolved to a Spanish news article on online bushfire misinformation.16 Highlighted in the co-domain bigraph are two zones of domains that appear mostly linked to one or the other of the Supporter and Opposer nodes, which are, again, appear polarised in the network. The domains in these zones appear aligned again with Opposers referring to domains hosting DEBUNKING URLs and Supporters referring to domains hosting NARRATIVE URLs. A few domains are referred to very frequently by individual nodes (visible as dark, large edges), and these are often social media sites, such as YouTube, Instagram and Facebook.
4 Inauthentic behaviour analysis
Phase 1 | Phase 2 | Phase 3 | Overall | |||||
---|---|---|---|---|---|---|---|---|
Count | % of All | Count | % of All | Count | % of All | Count | % of All | |
Supporters | ||||||||
All | 1573 | 121 | 5278 | 6972 | ||||
Hashtags | 1 | 0.1 | 0 | 0.0 | 19 | 0.4 | 20 | 0.3 |
Hashtags + URL | 160 | 10.2 | 7 | 5.8 | 502 | 9.5 | 669 | 9.6 |
Mentions + Hashtags | 60 | 3.8 | 3 | 2.5 | 277 | 5.2 | 340 | 4.9 |
Mentions + Hashtags + URL | 12 | 0.8 | 2 | 1.7 | 59 | 1.1 | 73 | 1.0 |
Opposers | ||||||||
All | 33 | 327 | 3227 | 3587 | ||||
Hashtags | 0 | 0.0 | 0 | 0.0 | 0 | 0.0 | 0 | 0.0 |
Hashtags + URL | 1 | 3.0 | 3 | 0.9 | 43 | 1.3 | 47 | 1.3 |
Mentions + Hashtags | 0 | 0.0 | 0 | 0.0 | 0 | 0.0 | 0 | 0.0 |
Mentions + Hashtags + URL | 0 | 0.0 | 0 | 0.0 | 5 | 0.2 | 5 | 0.1 |
Unaffiliated | ||||||||
All | 1961 | 759 | 14,267 | 16,987 | ||||
Hashtags | 2 | 0.1 | 0 | 0.0 | 32 | 0.2 | 34 | 0.2 |
Hashtags + URL | 181 | 9.2 | 14 | 1.8 | 434 | 3.0 | 629 | 3.7 |
Mentions + Hashtags | 35 | 1.8 | 8 | 1.1 | 137 | 1.0 | 180 | 1.1 |
Mentions + Hashtags + URL | 18 | 0.9 | 1 | 0.1 | 83 | 0.6 | 102 | 0.6 |
#ArsonEmergency
hashtag in the majority of them. In six of the tweets, other accounts were mentioned, including prominent Opposer and Unaffiliated accounts, perhaps in the hope that they would engage by retweeting and thus draw in their own followers.#ArsonEmergency
#EcoTerrorism
#ClimateChangeHoax
’, in that order (cf., Pacheco et al. 2021, case study 3), is another potentially simple yet informative analysis relying on sequence mining (Mooney and Roddick 2013).5 Discussion
#ArsonEmergency
was created specifically to counter #ClimateEmergency
(Graham and Keller 2020) and may even have been part of a broader disinformation campaign involving elements of the political and media elite (Keller et al. 2020). Aggressive language was observed in both affiliated groups, but troll-like tweet text patterns including only hashtags, mentions and URLs were employed far more often by Supporters, especially in Phase 3. Distinguishing deliberate baiting from honest enthusiasm (even with swearing) is non-trivial (Starbird et al. 2019; Starbird and Wilson 2020), but identifying targeted tweets lacking content is a more tractable approach to detect inauthentic and potentially malicious behaviour.5.1 A disinformation campaign?
#ArsonEmergency
was deliberately created (Graham and Keller 2020), forming a ‘data deficit’ (Smith et al. 2020) for the sharing of misinformation regarding the arson narrative. This could form an isolated echo chamber for recruiting a new user base and radicalising it. Then, once established, it could link into the broader discussions by using a variety of hashtags in their tweets, which is what we observed. Radicalisation may not have been the ultimate goal of this particular community, but the technique could equally be used to garner support. Large isolated communities of accounts have been discovered by researchers before,17 and moderate levels of activity could remain undetected, particularly if participants avoided using other hashtags in their #ArsonEmergency
tweets (which would link to other hashtag communities). #ArsonEmergency
was discovered because participating accounts were known to Graham and Keller . This study provides confirmation of the presence of trolling, but no direct evidence of disinformation (cf., Graham and Keller 2020; Keller et al. 2020).5.2 Strategies for countering misinformation
#ArsonEmergency
. Supporter numbers and activity rose dramatically after the story reached the MSM, drawing in many overseas contributors and shifting towards more inauthentic behaviour patterns. In contrast, the Opposer response was swift and simple, focusing on retweeting links to the ZDNet article and other fact-checks and official information, as it became available. Opposer activity was highest in Phase 2, but may have helped provide content for the incoming Unaffiliated accounts to share. In this way, the Unaffiliated accounts eventually shared DEBUNKING articles much more frequently than NARRATIVE aligned ones in the third phase. This occurred despite great increases in activity by Supporters, including relatively more uses of hashtags, mentions, replies, retweets and quotes than in Phase 1.#ArsonEmergency
);5.3 Recommendations for future studies
5.4 Methodological contributions
6 Conclusion
#ArsonEmergency
activity on Twitter in early 2020 provides a unique microcosm to study the growth of a misinformation campaign before and after it was widely known, forming a natural experiment. Here, we have shown that polarised groups can communicate over social media in very different ways while discussing the same issue. In effect, these behaviours can be considered communication strategies, given they are used to promote a narrative and represent attempts to convince others to accept their ideas. Supporters of the arson narrative used direct engagement to reach individuals and hashtags to reach groups with a wide range of URLs to promote their message, while Opposers focused on using retweets and a select set of URLs to counter their message. Supporter activities resulted in them being deeply embedded and distributed in the interaction networks, yet Opposers maintained high centrality and were supported by and appeared to coordinate with active Unaffiliated accounts. The counteraction appears to have been successful, with the predominant class of articles shared being shifted from narrative aligned in Phase 1 to debunking articles in Phase 3. Graham and Keller’s efforts to draw attention to the #ArsonEmergency
discussion (Stilgherrian 2020), and the subsequent associated MSM attention, is likely to have contributed to this effect, given the significant increase in discussion participants in Phase 3. This highlights the value in publicising research into misinformation promotion activities.