Measuring the Impact of ISIS Social Media Strategy

1 Measuring the Impact of ISIS Social Media StrategyMajid Alfifi, Parisa Kaghazgaran, and James CaverleeFred MorstatterDepartment of Computer Science & EngineeringInformation Sciences InstituteTexas A&M UniversityUniversity of Southern CaliforniaCollege Station, TXMarina Del Rey, CA{alfifi, kaghazgaran, groups like the Islamic State of Iraq and Syria (ISIS) haveexploited Social Media such as Twitter to spread their propagandaand to recruit new members. In this work we study the extent towhich ISIS is able to spread their message beyond their immediatesupporters. Are they operating in their own sphere with limitedinteraction with the overall community? Or are they well rootedamong normal users? We find that three-quarters of the interactionsISIS received on Twitter in 2015 actually came from eventuallysuspended accounts raising questions about the potential numberof ISIS-related accounts and how organic ISIS audience is.}

2 Towardstackling these questions, we have created a unique dataset of 17million ISIS-related tweets posted in 2015. This dataset is availablefor research purposes upon INTRODUCTIONThe past few years have seen Social Media as an effective toolfor facilitating uprisings and enticing dissent in the Middle East[4][11][14]. The embrace of Social Media in the region has made it abattle ground for ISIS (and similar groups) versus existing regimes,all spreading propaganda, recruiting sympathizers, and undermin-ing rivals [4]. Social Media has given terrorists the ability to directlycome into contact with their target audience and either spread ter-ror or recruit. In fact, ISIS has been repeatedly described as the mostadept terrorist group at using Internet and Social Media propagandato recruit new members [13]. Several studies have looked into howISIS operates on platforms like Twitter [3] [8] [12] but little hasbeen done to quantify their true reach and Impact .

3 While there aremany factors that could have contributed to their success such asgaining territories on the ground and appealing to the communitieswhere they operate [12] quantifying the Impact of their socialstrategy is still crucial for a better understanding of their addition, although Twitter suspends accounts that violate theterms of service, ISIS accounts seem to succeed in staying longeron the service posting more tweets than other malicious eventuallysuspended accounts (Figure 1) making studying their Impact evenmore we discuss in more detail in Section 3, several ISIS-relateddatasets have been made available to the research community. OurPermission to make digital or hard copies of part or all of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citationon the first page.

4 Copyrights for third-party components of this work must be all other uses, contact the owner/author(s).MIS2, 2018, Marina Del Rey, CA, USA 2018 Copyright held by the owner/author(s).ACM ISBN 123-4567-24-567/08 suspended accountsISIS accountsNumber of tweets before suspensionFraction of suspended usersFigure 1: ISIS accounts spread more content before gettingsuspended by Twitter compared with other eventually sus-pended complements those efforts by contributing a more compre-hensive dataset spanning the year 2015. Concretely, we share withthe research community a dataset of 24k ISIS users and their 17million tweets that span all of 2015. We additionally make an initiallook into the Impact of their behavior on the Twitter communityas a whole in Section DATASETWe have access to a large dataset of billion tweets representingall tweets generated in the Arabic language in 2015 through privatefull access to the Twitter Firehose.

5 In addition, we exploit a crowd-sourcing initiative by the Anonymous hacking group that invitedArabic speakers to report Twitter accounts that they think wereassociated with ISIS. This effort originally identified more than25,000 ISIS sympathizers and supporters through only use accounts that have actually been suspendedby Twitter, indicating multiple users have reported those accountsand that the Twitter spam control team found those accounts inviolation of the terms of service. We then look for their tweetsand their interactions with the Twitter community in our largerdataset. A previous work [8] used this same ISIS users dataset but,due to Twitter limitations, they were limited to 10% of the totaltweets through the Truthy project at Indiana University [6]. In1 The original website hosting these accounts has been taken offline but we were ableto recover accounts from ,88017,434,323 ISIS-Retweets551,86910,436,603 ISIS-Mentions745,72119,570,380 Legit-Tweets23,88017,454,068 Legit-Retweets1,753,19512,175,619 Legit-Mentions2,161,10617,479,990 Table 1: ISIS-Tweets are tweets posted by a known seed ofISIS-related accounts.

6 Legit-Tweets is a randomly sampledset of users and their tweets. Retweets and mentions of thesetwo sets (ISIS and Legit) by the overall Twitter communityare also , in this work we were able to recover all of the contentthese accounts generated in 2015 with our private full access to theTwitter Firehose. Note that this dataset contains only tweets in theArabic language but since ISIS is mainly active in the Middle Eastand the majority of its members are Arabic speakers, we believethis dataset provides a clear window into ISIS Social Media study the Impact of ISIS on the overall Twitter community, wefocus here on their interactions with the community. We considerretweetingas a signal that a user has seen the tweet although theymay or may not agree with it. We also considerreplyormentionof a user as an interaction event that often involves discussion,agreement or disagreement.

7 We measure the size of the interactionwhich is useful in understanding ISIS penetration in the Twittercommunity. In addition, we make a general observation about thepotential sentiment of those who interacted with ISIS based on thekeywords used [12].We select two groups of tweets one that represents ISIS-relatedtweets and another randomly sampled group that we use to compareISIS to other normal users. Concretely, we have the following setsof data (summarized in Table 1):ISIS-Tweets:Tweets posted by the reported ISIS-related are 23,880 accounts that generated 17,424,323 :All retweets of ISIS tweets including those fromthemselves. There are 10,436,603 retweets, posted by 551,869 :All tweets that mention any of the ISIS are 19,570,380 such tweets generated by 745,721 order to better understand the behavior and interactions ofISIS accounts, we randomly sampled an equal-size set of legitimateaccounts ( accounts that were still alive on Twitter by end of 2016)and use them for comparison.

8 Concretely we have the following:Legit-Tweets:A random sample of 23,880 legitimate accounts posted 17,454,068 :All retweets of the above legit tweets. We found12,175,619 retweets generated by 1,753,195 :All tweets mentioning the above legit tweets. Wefound 17,479,990 tweets generated by 2,161,106 RELATED WORKIn this section, we first discuss datasets that are relevant to mea-suring the Social Media communications of extremists and thenwe introduce other approaches that have investigated the onlinecommunication of and Morgan identified 46,000 ISIS supporter accounts onTwitter [4]. They began with 454 accounts known as ISIS supportersas initial seeds. They then collected all accounts following thoseseeds. This approach continues to two further steps on the followed-by network. They introduced a classification task to determine if auser is an ISIS et al.

9 , collected 3 million Arabic tweets referring to ISIS[12]. They classified users based on how they call ISIS. The full nameof Islamic State is the indicator of support, while abbreviationslike ISIS or Daesh indicate opposition. Zaman [15] introducedan ISIS dataset consisting of about 17,000 tweets from about 100 ISIS supporter accounts. Keywords, Images, and network-basedfeatures were used to classify a user as pro-ISIS or another effort, Bodine-Baron et al., collected 23 million tweetsthat referenced the Arabic versions of either Islamic State or Daesh [5]. This corpus consists of both ISIS supporters and op-ponents whom they were separated by which phrase they use todescribe ISIS: ISIS followers refer to ISIS as the Islamic State whiledetractors often use the term Daesh . They lexically analyzed thetweets and found that users saying Daesh use other terms suchas Terrorist , Kharijites , militants , dogs of fire , and dogs ofBaghdadi that prove they are highly critical of ISIS.

10 On the otherhand, tweets with Islamic State contain glowing terms such as monotheists Mujahideen , Soldiers of the Caliphate , and lions ofthe Islamic State .In a recent effort, Badawy et al., investigated how ISIS supporterstake advantage of Social Media to spread their propaganda andrecruit militants by studying million tweets posted by 25,000accounts recognized as pro-ISIS and suspended by Twitter [2].The Islamic State heavily relies on Social Media for there has been work that focuses on analyzing the datamade by this group. For example, Farwell [7] studied the mediastrategies of ISIS. They described two conflicting strategies fol-lowed by ISIS: they try to protect the identity and location of theirleaders by minimizing Internet communications while they takeadvantage of Social Media for recruitment. Gates and Potter [9]analyzed who creates the content of the organization s recruitmentmaterials and found that they use a globally distributed network ofvolunteers who create content to fit the aesthetic of their particularregion.