#FailedRevolutions: Using Twitter to study the antecedents of ISIS support
First Monday

#FailedRevolutions: Using Twitter to study the antecedents of ISIS support by Walid Magdy, Kareem Darwish, and Ingmar Weber



Abstract
Lately, the Islamic State of Iraq and Syria (ISIS) has managed to control large parts of Syria and Iraq. To better understand the roots of support for ISIS, we present a study using Twitter data. We collected a large number of Arabic tweets referring to ISIS and classified them as pro-ISIS or anti-ISIS. We then analyzed the historical timelines of both user groups and looked at their pre-ISIS period to gain insights into the antecedents of support. Also, we built a classifier to ‘predict’, in retrospect, who will support or oppose the group. We show that ISIS supporters largely differ from ISIS opposition in that the former referred a lot more to Arab Spring uprisings that failed than the latter.

Contents

Introduction
Background
Data collection
Classifying and analyzing pro/anti-ISIS Twitter users
Discussion
Conclusion

 


 

Introduction

The recent rise and territorial gains of the “Islamic State of Iraq and Syria” (ISIS) (or simply the “Islamic State”) have sparked significant interest in the group. One particular aspect of interest in ISIS is related to the profile of individuals who are likely to join or otherwise support them. Since the events of September 11, there has been wide interest in identifying necessary and sufficient traits individuals may possess that are likely to propel them to support or join violent militant organizations. Contrary to popular belief, social psychology studies and reviews suggest that individuals who end up joining such organizations are typically more educated, financially better off, generally more accomplished, and more exposed to Western culture than average (Louis, 2009). Many studies have looked at whether such individuals suffer from psychological disorders, but found no evidence of such (Horgan, 2003). On the contrary, these individuals generally exhibit higher than average psychological strength and are far less traumatized by incarceration, interrogation, or imprisonment than the ordinary civilian (Miller, 2006). In other words, they are not just normal people, but rather they are normal people with better than average fortunes. Since millions of people fit this description, and the overwhelming majority of them do not resort to violence, psychological traits that were studied are insufficient identifiers of people who are inclined to join terrorist organizations.

In a study of a leftist group in India which resorted to violence (Sarangi and Alison, 2005), the 12 members of the group who were interviewed provided similar personal narratives in which they described structural deficiencies in their societies, including corruption, oppression, and lack of empowerment for certain segments of the society. Their personal narratives painted a picture of those engaged in violence as “heroes” who were confronting these deficiencies. The findings of Sarangi and Alison (2005) are in line with the findings of many other studies that show that people who resort to anti-establishment violence do so to eliminate what they perceive as injustices (Plous and Zimbardo, 2004). Along the same lines, there was a story concerning a man called Ahmed Al-Darawy, a successful and affluent 38 year-old former manager in a multinational company in Egypt, who was killed while fighting alongside ISIS [1]. Al-Darawy was a former police officer who left the service to join the revolutionaries who toppled the former Egyptian president Hosni Mubarak, and he later ran for elected office in an embrace of democracy. Interviews with his friends after his death painted a picture of a person who was disillusioned and angry, particularly after the military intervention against the democratically elected president, Mohamed Morsi. Al-Darawy exemplifies the previous description of people who resort to violence. Juergen Todenhoefer, a German journalist who recently spent 10 days with ISIS, provided confirming testimony in which he said “These [ISIS fighters] are not stupid people. One of the people we met had just finished his law degree, he had great job offers, but he turned them down to go and fight.” [2] There is evidence in the literature that support for such militant groups is mostly political, rather than religious (Abdulla, 2007). In short, Louis [3] attributes the motivation to pursue violent means to achieve political ends to the following beliefs: 1) “Alternatives to terror do not work”; and 2) “Terrorism can achieve social change”. Along the same lines, some studies on social movements have concluded that political repression may lead people to consider violence as a justified means to further their goals (Della Porta, 2008).

Since playing an important role in the Arab Spring, Twitter has established itself as an important communications medium for grassroots discourse surrounding various political topics, including ISIS. In this paper, we identified Arab Twitter accounts with explicitly expressed positions supporting or opposing ISIS. We then used these accounts to identify distinguishing features from the users’ Twitter profiles that foretell their future positions prior to authoring the first tweet overtly declaring their stance. Understanding these distinguishing features can shed light on why certain users later support ISIS. We examine the interests of both groups before and after they started explicitly supporting or opposing ISIS. In line with the findings in the literature, we attempt to analyze the motivations that prompt people to express support for ISIS (albeit not necessarily join them). In particular, we are looking for evidence on whether perceived injustices act as triggers.

We conducted the study on the accounts of 57,000 Twitter users who authored tweets about ISIS between 13 October and 1 November 2014. Given these user accounts, we crawled their timelines, from which we collected nearly 123 million tweets, representing their historic tweets including the period before they started posting about ISIS. The contributions of this paper are as follows:

  • We determine distinguishing language that signals current support for or opposition to ISIS. As shown later, the overwhelming majority of users are not neutral towards ISIS.
  • We train a classifier that can predict future support for or opposition to ISIS with 87 percent accuracy.
  • We show the differences in interests between user groups who subsequently support or oppose ISIS.
  • We introduce a methodology which can be applied to study antecedents of other cases of polarization.

 

++++++++++

Background

ISIS and the Syrian conflict online

Militant groups have long used the Internet and social media for communication, information gathering, recruitment, and social network engagement (Cohen-Almagor, 2012). Although specific literature on the use by ISIS of social media is scant, there are strong indicators that the group seems to be fairly media savvy with strong presence on different social media platforms (Shane and Hubbard, 2014; Allendorfer and Herring, 2015). For example, ISIS (or its affiliates) maintain many Twitter accounts that propagate the group's message in several languages. When accounts are taken down, new ones are quickly created to replace them. ISIS also produces high quality videos in several languages as a recruitment tool. For example, the group produced an English video entitled “There is No Life Without Jihad” that features Western fighters who exhort others to join them (Kohlmann and Alkhouri, 2014). Much of the propaganda of ISIS is focused on reporting military successes and ISIS’ success in implementing different aspect of Sharia (Islamic law). Although it was not the first to engage in this kind of media campaign, it has eclipsed other militant groups, such as Al-Qa’ida, in attracting new members. Its recruitment success can mostly be attributed to their military victories, particularly the capture of Mosul, Iraqvs second largest city. Kohlmann and Alkhouri (2014) provide accounts of several individuals who either joined or attempted to join ISIS or other militant groups in Iraq and Syria. In most accounts, individuals expressed strong anti-American sentiment, including one fighter who was filmed tearing up his American passport and burning it, another who said that killing American soldiers was justified, and yet another who spoke of a war that the West was waging against Muslims. Disdain for US intervention in the Middle East seems to be a common theme among Arabs in specific and Muslims in general (Jamal, et al., 2015). Klausen (2015) examined the tweets of 59 Westerners who are believed to be fighting alongside ISIS. The analysis shows that most of the content of the tweets is dedicated to spreading religious material relating to Jihad (39 percent) or reporting from the battle field (40 percent). Klausen notes that there seem to be some “feeder” accounts that control the messages of some of these accounts. Farwell (2014) examined the ISIS social media strategy and its effectiveness. He states that the strategy of ISIS is intended to intimidate enemies, recruit new members, and also paint a human image of its members with their pictures of their members “eating Snickers” and playing with kittens.

A recent report by the Brookings Institution examined 20,000 Twitter accounts that expressed support for ISIS between September and December 2014 (Berger and Morgan, 2015). The report estimates that there are at least 46,000 supporting accounts, with a core number of 500–2,000 hyperactive accounts, despite the fact that Twitter is actively suspending accounts that support ISIS. Most of the accounts were recently created (mostly in 2013 and 2014). The breakdown of languages used in these accounts is 75 percent Arabic and 20 percent English, with the remaining in other languages. In our study, we focused exclusively on tweets in Arabic.

Further, U.S. military intervention in Iraq and Afghanistan seems to coincide with a sharp increase in terror attacks in both countries. Figure 1 shows a steady rise in terrorist attacks in Iraq starting in 2003 (the year that the U.S. intervened in Iraq). In most cases, individuals joining ISIS obtained information from the internet in the form of tweets from notable individuals, personal contact over social media, or online videos and lectures. Thus, there are indications that the media strategy of ISIS is effective in attracting supporters (Kohlmann and Alkhouri, 2014). The effectiveness of the social media strategy of ISIS and the boost it received due to the bombing campaign against it has been reported on in the news [4].

 

Number of terrorist attacks in Iraq per year
 
Figure 1: Number of terrorist attacks in Iraq per year (Source: Global Terrorism Database).

 

Although not explicitly focused on ISIS, other studies have looked at Twitter communication in Syria during the civil war. O’Callaghan, et al. (2014) performed a network analysis of 652 Twitter accounts of Syrians engaged in the conflict. These accounts were identified using Twitter lists [5], and their analysis revealed four general groups, namely: (i) pro-Assad, (ii) Kurdish, (iii) secular/moderate opposition, and (iv) Islamists, including supporters of ISIS. Their analysis, performed before the major territorial gains of ISIS of mid-2014, also involved content analysis of the Freebase [6] concepts assigned to YouTube videos that were uploaded by the Twitter users. Our analysis takes a different approach and identifies a wider range of users (57k, extracted from an initial set of 165k) and, more importantly, goes back in time to try to understand the antecedents of joining or merely supporting a particular group. Another study investigated the reactions of Twitter users who mentioned in their tweets the 2013 sarin gas attacks near Damascus in the two days following the attack (Tyshchuk, et al., 2014). Among other things, the authors observed that “there was no immediate polarization of opinions following the event”, and that “Twitter communities were too sparse to produce substantial amount of social pressure to force an opinion/opinion shift”. Finally, work by Morstatter, et al. (2013) is related as they used geo-tagged tweets from Syria as a data source. Their analysis, however, evaluates the bias between different sources and sampling methods of Twitter data and does not look at ISIS at all.

Using online data to study political polarization

Our pro- vs. anti-ISIS angle can also be seen as a study of online polarization. Polarization, usually along the political left vs. right dimension, has a long research history. One quantitative approach is the NOMINATE score (Poole and Rosenthal, 1985), which uses data of roll call [7] votes from the U.S. Senate and House of Representatives to quantify the statement that U.S. politics follows a one-dimensional left-to-right schema. With a focus on online data, polarization in U.S. politics has been studied in the setting of blogger networks (Adamic and Glance, 2005) and Twitter (Conover, et al., 2011; Golbeck and Hansen, 2011; to name two influential studies). Geographically and topically closer to ISIS, online polarization and political violence has been studied using Twitter data from Egypt (Borge-Holthoefer, et al., 2015; Weber, et al., 2013). Other work by Al-Ani, et al. (2012) looked at Egyptian blogs from the seven-year period leading up to the 2011 Egyptian uprising that challenged the government’s narrative. The conflict between Israel and Palestine and how it plays out on Twitter has also recently been studied (Liu and Weber, 2014). Although these studies describe differences in the users who are part of two opposing camps, aside from the work of Al-Ani, et al. (2012), none of these studies look at the antecedents or attempts to explain what makes a user join or support one side rather than another.

 

++++++++++

Data collection

Though ISIS attracts supporters from many different countries where many languages are used (Berger and Morgan, 2015; Allendorfer and Herring, 2015), the majority of ISIS supporters come from the Arab world (Berger and Morgan, 2015). Hence, we focused our study on Arabic Twitter profiles. Figure 2 summarizes the steps of our data collection. We started by collecting tweets mentioning ISIS, then we collected the timelines of the Twitter accounts that authored/retweeted those tweets and applied our analysis to the group of accounts that showed clear interest in the topic.

 

Steps performed for data collection
 
Figure 2: Steps performed for data collection.

 

Collecting and classifying tweets on ISIS

We start by identifying Arabic tweets by continuously searching the Twitter Rest API with the query “lang:ar” [8]. From this set of Arabic tweets, we then collected all tweets mentioning ISIS by any of its name forms between mid-October and end of December 2014. The name variations were of two types, namely: those that used the full name of the group, such as “الإسلامي الدولة” (Aldawla Alislamiya — “Islamic State”) or “الإسلامية في العراق والشام” (Aldawla Alislamiya fi Aliraq walsham — “Islamic State in Iraq and the Levant”), and those that used an abbreviated version of the name, such as “داع” (da’esh — Arabic acronym for the group), “عش” (da’eshy — from da’esh), or “دواع” (dawa’esh — plural of da’eshy).

In all, we collected 3.9 million Arabic tweets mentioning ISIS in some form between 13 October and 31 December 2014. Using these tweets, we were interested in distinguishing between Twitter users who support ISIS and those who oppose ISIS.

There have been several reports from CNN (Sanchez, 2015), the Independent (Dearden, 2014), and International Business Times (Ross, 2014) claiming that supporters of ISIS tend to use the full name of the group, while those who oppose it tend to use an abbreviated form. We set out to quantitatively validate this claim. To do so, we randomly selected 1,000 tweets, sampled uniformly across all days, containing the full name of ISIS, and another 1,000 tweets containing any of the abbreviated forms of the group. A native Arabic speaker who is well acquainted with the topic was asked to label the tweets as pro-ISIS, anti-ISIS, or neutral. Pro-ISIS tweets included ones that praise the group, spread its messages, or report on it positively. Anti-ISIS tweets included ones that attack the group or spread negative news about it. Neutral tweets included those that merely report news about ISIS or spam tweets that use ISIS-related hashtags to increase their exposure.

Figure 3 presents the percentage of tweets that support, oppose, or are neutral towards ISIS based on the use of: a) the full name (or its variants); or b) the abbreviated name (or its variants). The results confirm the claims in Sanchez (2015), Dearden (2014), and Ross (2014). As shown, using the full name of the group is a strong indicator of support for ISIS (93 percent), while the probability of using the full name to show opposition was only 1.2 percent. In contrast, using an abbreviated form is a general indication of opposition (77 percent). However, an abbreviated form is still used by supporters 7.5 percent of the time, and 15 percent of tweets using an abbreviated form of the name reveal no specific leaning. This makes the abbreviation a weaker indicator of opposition compared to the full name as an indicator of support. However, both could be used as effective features for classifying accounts.

 

Percentage of pro-/anti-ISIS tweets given the presence of either an abbreviated form or the full name in Arabic
 
Figure 3: Percentage of pro-/anti-ISIS tweets given the presence of either an abbreviated form (e.g., “ISIS”) or the full name (e.g., “Islamic State”) in Arabic.

 

Based on this observation, out of the 3.9 million tweets that we collected, an estimated 1.8 million and 1.7 million tweets were supporting and opposing ISIS, respectively, while the rest (0.4 million tweets) were estimated to be neutral, which account for roughly 10 percent of the tweets.

Collecting and classifying Twitter accounts interested in ISIS

From the collected tweets, we identified 180k accounts with at least one tweet mentioning ISIS during the period between 13 October and 1 November 2014. We crawled their timelines to conduct user-level analysis later. We obtained the historical tweets of these accounts through the Twitter Rest API. For all accounts, we downloaded their last 3,200 tweets.

Out of the 180k accounts, only 165k had active accounts by the time we finished crawling on 17 November 2014, while the remaining 15k accounts had either been deleted by the users, been suspended by Twitter, or had their privacy settings changed to “protected”. The closing of a large number of pro-ISIS accounts is consistent with observations made by the Brookings Institution study (Berger and Morgan, 2015). A total of 324 million tweets were collected, along with other Twitter profile information such as profile creation date and user declared location; see Figure 2.

The next step was to confidently identify user accounts that support or oppose ISIS. Since the use of either the full or abbreviated names of ISIS is mostly (but not completely) indicative of support (93 percent) or opposition (77 percent), respectively, we elected to focus on Twitter users who authored a certain minimum number of tweets, with most of them including either the full or the abbreviated form of the name. This would provide greater classification confidence. When we set out to identify accounts with a minimum number of tweets about ISIS, we noticed that more than 16k accounts did not have any tweet about ISIS in their timelines. This was surprising, since these accounts were considered because they had at least one ISIS related tweets in our initial set of 3.9 collected tweets. The reason behind this could be that these users either deleted the tweets about ISIS, or these tweets were retweeted from other accounts that deleted the tweets or were suspended. This led to the presence of only 148k accounts with at least one tweet on ISIS.

Figure 4(a) plots the distribution of 148k accounts based on the number of tweets mentioning ISIS in their timelines, and the percentage of accounts that would remain if those with a small number of tweets were filtered out. As shown, approximately 50 percent of the accounts mentioned ISIS fewer than 10 times in their timeline, and we decided to discard these accounts. The remaining 63k accounts corresponded to users actively engaging with the topic.

 

a Distribution of accounts having Ntweets on ISIS (black line, axis on left), and the percentage of accounts remaining after filtering out accounts with small Ntweets (brown line, axis on right); b Histogram of accounts according to the degree of polarization of using one type of mention of ISIS over the other
 
Figure 4: (a) Distribution of accounts having Ntweets on ISIS (black line, axis on left), and the percentage of accounts remaining after filtering out accounts with small Ntweets (brown line, axis on right). (b) Histogram of accounts according to the degree of polarization of using one type of mention of ISIS over the other.

 

To determine the leaning of each of these accounts, we computed the degree of polarization of each account by computing the percentage of using one form of ISIS reference (full name or abbreviation) vs. the other. Figure 4(b) shows that 56 percent of Twitter users strictly use either the full or abbreviated name forms of ISIS in over 90 percent of their tweets about the group. Further, 90 percent of the users use one form in over 70 percent of their tweets that mention the group. Based on these observations, and in an effort to achieve high classification precision, we retained the 57k Twitter accounts that use one form of the name at least 70 percent of the time and that have authored at least 10 relevant tweets.

Table 1 summarizes all the stages of filtering the collected accounts with the exact numbers of accounts in each stage. Of the remaining 57k accounts, 11,332 were automatically labeled as pro-ISIS as they predominantly used the full name, whereas 45,628 were labeled as anti-ISIS as they predominantly used the abbreviated form. The number of tweets collected for these users was 123 million tweets. An interactive tool to explore this data set was recently presented by Magdy, et al. (2015).

 

Table 1: Summary of the number of accounts at each stage of filtering to reach the final set of accounts used in the study.
Number of accountsDescription
180,431unique accounts initially extracted from tweets
165,301active accounts by end of crawling time
148,753at least one tweet on ISIS in timeline
63,305at least 10 tweets on ISIS in timeline
56,960at least 10 tweets and 0.7 polarization

 

To assess the accuracy of our automatic labeling, we randomly selected 200 Twitter accounts, half of which were labeled as pro-ISIS and the other half as anti-ISIS. For manual assessment, we enlisted the help of two annotators other than the one who labeled the initial set of tweets, so as to avoid potential labeling bias. Out of the 200 accounts to be labeled, 50 accounts were common between the two annotators to measure inter-annotator agreement. All tweets mentioning ISIS in the timeline of the 200 accounts were provided to annotators. Based on the manual annotation, the accuracy of automatic labeling is 98 percent [9]. The inter-annotator agreement was 96 percent (Cohen’s kappa, κ = 0.92). This shows that the simple method of using name mentions of the group for classifying Twitter users to be supporting or opposing ISIS is highly accurate.

 

++++++++++

Classifying and analyzing pro/anti-ISIS Twitter users

Classifying Twitter users

Given the users we have identified as being most likely supporting or opposing ISIS, we were interested in classifying users as potentially supporting or opposing ISIS before they explicitly write a tweet identifying their stance. As mentioned earlier, we automatically labeled 11,332 and 45,628 accounts as supporting or opposing ISIS. Of the 11,332 supporting accounts, 4,307 indicated support for ISIS starting from the very first tweet we could obtain. As this user group did not have any pre-ISIS data, we could not use them for prediction, leaving us with 7,225 accounts. For balance during classification, we randomly picked an equal number of anti-ISIS users who were active prior to their opposition to ISIS. We used these accounts with the tweets they authored prior to the first supporting/opposing tweets for classification. We randomly selected 10 percent for parameter tuning and performed 10-fold-cross-validation on the remaining 90 percent. We trained a Support Vector Machine (SVM) classifier using the SVMLight implementation with a linear kernel. The threshold was adjusted to maximize the classification effectiveness (F1 score). For features, we used bag-of-words features, including individual terms, hashtags, and user mentions. The Arabic text of the tweets was processed using the method described by Darwish, et al. (2012) to normalize the text and to handle word elongations. We combined all the tweets prior to the first tweet mentioning ISIS for a single user into one document, and we used all the tokens in a combined-tweets document as features. Table 2 lists classification results of potential supporters and opponents of ISIS based on the tweets they authored prior to their first tweet explicitly stating their position. As the results show, both groups are quite distinct and hence separable from each other even before they voice explicit support or opposition, and the classifier can distinguish between them with relatively high accuracy.

 

Table 2: Classification results of potential ISIS supporters/opponents.
 PrecisionRecallF1-measure
Pro-ISIS89.683.786.6
Anti-ISIS84.790.387.4
Average87.287.087.0

 

Next, we were interested in understanding the underlying features that make the two groups separable. So, we consulted the SVM classification model to identify the most distinguishing 50 hashtags that the classifier used to determine if a person would potentially support ISIS and the most distinguishing 50 hashtags that would indicate future opposition. We opted to use hashtags instead of individual words for this exploration, because they are more demonstrative of underlying topics. Furthermore, we found that 34.3 percent of the tweets contain hashtags, which is significantly higher than the global Twitter average, which ranges from four percent to 25 percent according to the language (Weerkamp, et al., 2011). This would be sufficient to model the discussed topics effectively.

Figure 5 shows tag clouds for the hashtags, mostly in Arabic, for future pro- and anti-ISIS users in the before condition. The size of each hashtag in the cloud is indicative of its weight as assigned by the classifier.

 

Tag clouds for the most indicative 50 hashtags of ISIS support and opponents in the period before and after they started tweeting about ISIS
 
Figure 5: Tag clouds for the most indicative 50 hashtags of ISIS support and opponents in the period before and after they started tweeting about ISIS.

 

By examining the top 50 hashtags that are most indicative of future ISIS support, we observe the following (each hashtag is followed by its rank on the list):

  • Most hashtags are political and are specifically related to support for the Arab Spring and opposition to regional regimes. These hashtags by country are as follows (translated to English):
    • Egypt: #elect_the_pimp (4 — hashtag referring to Sisi, the current Egyptian president, that emerged during the Egyptian presidential elections), #Morsi (7 — ousted Egyptian president), #Rami_Lakkah (13 — Egyptian politician and businessman), #Rabia_Adawiyya (29 — sit-in that was disbanded violently by security forces), #Rabia (35), #the_free_25 (47 — referring to the Jan. 25 uprising)
    • Saudi Arabia: #Alburayda_sit-in (5 — protesting the incarceration of activists), #Saud_clan (9 — Saudi rulers), #the_people_say_their_word (18), #finishing_off (23 — opposition campaign), and #million_man_march_for_Haila_Qusair (32 — opposition campaign)
    • Iraq: #Iraqi_Spring (12)
    • Libya: #Libya (6), #Feb17 (17 — start date of 2011 revolution), #Gaddafi (45)
    • Syria: #Annusra_Front (19 — rebel group), and #Levant (25)
    • Kuwait: #National_dignity_march (24 — protest movement in 2012)
    • Jordan: #Jordan_star (31 — protest movement)
  • Some hashtags relate to conflict areas in the Muslim world, namely: #ajagaza (1 — Aljazeera Arabic, Gaza), #Mali (40), and #Qassam (44 — Hamas’ military wing)
  • A few hashtags relate to civil-rights or its violation, namely: #incarceration (11, 43) and #prisoners (41). This is in line with previous research that shows that imprisonment and torture seem to “fan the flames” of violence (Bellamy, 2009).
  • Only three hashtag are indicative of anything religious, namely #tweet_Islamic (16), #Islam (34), and #remembrance (49)
  • One interesting hashtag, #a_million_atheist_Arabs, was used for discussions with atheists.

In short, most of the hashtags are related to support for the Arab Spring, opposition to existing regimes in the Middle East, solidarity or concern over hotspots in the Muslim and Arab worlds, and disillusionment with the current status quo.

As for the top 50 hashtags that are most indicative of future opposition to ISIS, we observe that:

  • Most of the tweets are general, like the one about an exhibition in Abu Dhabi (1), #Instagram (16), #Reem_baklava_from_Istanbul (29), etc.
  • Some hashtags express political views, namely:
    • support for the coup in Egypt: #long_live_Egypt (9 — slogan of pro-Sisi supporters) and #our_country_movement (30 — a pro-Sisi movement)
    • support for UAE's political leadership: #UAE_international_performance (32), #civil_service (35), #Mohamed_bin_Zayed_among_best_Islamic_leaders (44)
    • support for Gaza, which is the only sentiment in common with the other group: #Gaza (13 and 50), #Gaza_under_attack (20), and #I_am_with_Hamas (42)
    • Animosity towards the Muslim Brotherhood: #Brotherhood

As can be seen, latter hashtags that are most predictive of future animosity towards ISIS are not indicative of any political stances except support for the regimes in UAE and in Egypt. Members of this group share a concern with the future supporters of ISIS in their support for Gaza.

Whereas the previous analysis was done on classifiers built before users expressed support or opposition to ISIS, we also built a separate SVM classifier exclusively using data after users expressed their opinion. Although this does not directly reveal anything about the antecedents of support, it still helps paint a clearer picture of how the supporters differ from the opposition. For this analysis, tweets containing a reference to ISIS were ignored because, trivially, otherwise the corresponding hashtags would have been the most discriminating features. Figure 5 in the after condition shows tag clouds for the most distinguishing hashtags, mostly in Arabic, for current pro- and anti-ISIS users, respectively. Again, we obtained the hashtags from the SVM classification model.

By examining the top 50 hashtags that are most indicative of current ISIS support, we note the following:

  • Similar to the hashtags that the Twitter users used before declaring their support for ISIS, many of the hashtags express animosity towards different regimes in the Middle East. Some of these hashtags by country are:
    • Kuwait: #Bodoon_schools (2 — Bodoon are persons without nationality), #Bodoon (33), #join_the_well (37 — protest movement)
    • Saudi Arabia: #the_people_say_their_word (3), #Saud_clan (14), #kidnapping_of_Saudi_women (20 — about government incarceration of women), #trial_drama (35 — mocking trials of opposition figures), #kidnap_of_May_Altalq_and_Amina_Alrashed (42 — same as #20), #salary_not_enough (50 — opposition campaign)
    • Egypt: #Rami_Lakkah (4), #Muslim_youth_uprising (8 — a day of protest in Egypt), #elect_the_pimp (12), #Sisi (15), and #Rabia (21).
    • ISIS-specific hashtags, including:
    • Media releases: #Despite_the_disbelievers (6) and #charge_of_Ansar (49 — about attack in Sinai against Egyptian army)
    • Other names for ISIS: #the_state (7 and 47), #Islamic_caliphate_state (13), #Islamic_caliphate (30), #State_of_Islam (32), and #Caliphate_state (41)
    • Battle-related: #Ein_Alarab (18 — Arabic name for Kobani, a town witnessing ongoing battles), #Ein_Islam (22 — ISIS’ name for Kobani)
    • ISIS media organizations: #Alfurqn_Foundation (28), #Islam_Post (32)
  • Another carry-over hashtag from the before condition is #incarcerated (23).

Among the hashtags indicative of opposition to ISIS, there seem to be three main themes, as follows:

  • The first theme includes supportive references to various rebel or Islamic groups in Syria such as #Alnusra_Front (2), #Islamic_Front (4), #martyrdom_of_Ahrar_Alsham_leaders (9), #Army_of_Islam (12), #Brotherhood_in_Syria (15), #Ahrar_Alsham (35). This can be attributed to ISIS attacks on opposition groups in Syria
  • The second theme includes:
    • support for some Middle Eastern regimes, namely United Arab Emirates, as in #Zayed_cultural_exhibition (20 — Zayed is the UAE founder), #love_UAE_and_its_leaders (27), #this_is_UAE (28), #we_love_Bu-khaled_the_Arab_leader (44 — Bu Khaled is vice president of UAE), Egyptian military regime as in #Egyptian_army_are_men (8), #long_live_Egypt (16), and #our_country_movement (39), and Saudi Arabia as in #Twitter_army (18) and #gunfire_in_Ahsaa (36 — an attack in Eastern Saudi Arabia)
    • opposition to various Islamic groups, as in #female_organization (14) and #Brotherhood_of_Syria and ISIS as in #khawarej (37 — a deviant sect in Islamic history) and #Baghdadi (42 — leader of ISIS).
    • opposition to regimes that are perceived to support Islamic groups, such as #Zaman_Arabic (5 — opposition publication in Turkey) and #why_oh_Qatar (19)
  • The third theme seems to be Shia related, as in #Hussein_lives_on (34 — revered Shia historical figure).

Example tweets and users

To provide a better qualitative understanding of our data set, we provide some pairs of before/after tweets for ISIS supporters (Table 3) and ISIS opposition (Table 4).

 

Table 3: Before (light gray) and after (white) tweets from Twitter users who end up supporting ISIS.
UserDateTweet (translated)
T125 May 2012Don’t be surprised if it rains today ... martyrs are spitting on us
9 November 2014Preliminary schizophrenia: I like ISIS, but I want to watch Chris Nolan’s new movie
17 November 2014The gazes of Bashar’s soldiers before slaughter by #Islamic_State in #despite_the_disbelivers
T226 August 2013Best pics from the liberation of Khanaser (Syrian town) #Syria #Alnusra_Front #Ahrar_Alsham
14 November 2014The strike of the Islamic State’s lions, Sinai State
T39 April 2014#mb_europe #elect_the_pimp Some of the military coup crimes in #Egypt (link)
17 November 2014Praise be to Allah, support for #Islamic_State in Indonesian mosques
T424 February 2011JUST IN: Benghazi cleaning the streets! (link) #Libya #GaddafiCrimes #feb17
25 September 2014Pictures of the dead of Crusader Arab alliance against Islamic State (link)
T59 March 2013An important message from your brother in Syria #jihad #Alnusra_Front #Free_Syrian_Army (link)
19 February 2014Islamic State retakes Babila (Syrian town) after #Free_Syrian_Army betrayal

 

 

Table 4: Tweets from anti-ISIS Twitter users after showing opposition to ISIS.
UserDateTweet (translated)
T612 November 2014Don’t be surprised if it rains today ... martyrs are spitting on us
15 November 2014Baghdadi state media #ISIS not different from Bashar state media, all in deceitful swamp
T79 September 2014Saddened for loss of heroes #martyrdom_of_Ahrar_Alsham_leaders. Same George Bush strategy: long dirty war
30 October 2014ISIS an internal Muslim problem, to be fixed by Muslim hands, not Americans or their Arab puppets
T817 August 2014Alnusra Front: pictures of your brothers in #resilient_Murek (Syrian city) (link)
23 August 2014Disgusting ISIS show to sell their pathetic goods (link)
T91 September 2014Hillary Clinton agrees with Muslim Brotherhood to establish ISIS #arrest_ISIS_Saudi_popular_demand
14 October 2014Why the dead children! Muslim Brotherhood and ISIS 2 sides of 1 coin #criminal_brotherhood #long_live_Egypt #mb_uk
T1024 October 2014Don’t let ISIS distract you from worship. This is their goal. #Hussein_lives_on
27 October 2014From the battlefield against ISIS #Hussein_lives_on (pic)

 

Among the ISIS supporters, user T1 seems to have supported the 2011 Egyptian revolution and is probably not an Islamist. User T3 opposed the military coup in Egypt. Users T2 and T5 supported different militant groups in Syria and ended up supporting ISIS. In the case of user T5, he transitioned from supporting to opposing the Free Syrian Army. User T4 was a supporter of the 2011 Libyan revolution. These examples show that the support to ISIS is not arising from ideological alignment as much as interest alignment. All of these examples shows that those users were supporting the Arab spring at some point in time, and ended up supporting ISIS, perhaps hoping that they may see ISIS fight dictatorial regimes after the failure of the Arab spring in most countries.

Among the ISIS opposition, users T6, T7, and T8 supported different Muslim rebel groups in Syria and ended up opposing ISIS. This is different from T5 who was supporting the same rebel groups in Syria, but ended up calling them traitors after he started supporting ISIS. User T9 supports the current Egyptian regime and opposes the Muslim Brotherhood. User T10 is seemingly a Shia fighting against ISIS.

 

++++++++++

Discussion

Methodological contribution

The method presented in this paper sheds light on the antecedents of ISIS support through an analysis of Twitter data. However, our method is in no way ISIS-specific, nor even Twitter-specific. It can be abstracted in the following way:

  1. Identify a research question where it is of interest to understand current and observable differences between groups in the light of not-so-distant events. Example: Given current expressions of support or opposition to ISIS, what are the antecedents of these opinion differences, rooted in past events?
  2. Identify an “always on” source of historical user data relevant to addressing the research questions. Example: Public Twitter data collectable via the REST API.
  3. Determine user groups of interest that differ in their current behavior. Example: ISIS supporters or opponents, identified by how they refer to ISIS.
  4. Choose a feature set that has relevance to the research question and is likely to differentiate between the groups of interest. Example: Counts of terms and hashtags.
  5. Use historical data to train a classifier with the training labels derived from current data. Example: Using a term-based, linear SVM classifier classify current ISIS supporters vs. opponents using their historical tweets.
  6. Perform a feature analysis to understand what differentiates the different groups. Example: Study the highest and lowest weights in the trained SVM model.
  7. Using a mixed methods approach, relate the feature analysis back to the research question. Example: Derive semantic clusters of related hashtags that are linked to past events and use usage differences of these clusters to understand current group differences.

Recently, Budak and Watts (2015) presented a research methodology they call ex-post panel construction. They note “that the ‘always on’ nature of Twitter allows researchers to construct ex-post panels of the sort we have introduced here months or even years after the events of interest have taken place, making it especially attractive for studying events such as political uprisings that are hard-to-impossible for researchers to anticipate and so do not lend themselves to traditional (i.e., ex ante) panel designs”. Their method differs from ours in that they are not using a classifier but, rather, are looking at directly compiled summaries of behavioral differences.

Our data collection involved using both the Twitter streaming API and, after identifying tweets and users of interest, the Twitter REST API to obtain historical information. However, due to the small temporal difference between when these two data sets were collected, some accounts had been deleted by the user, had their privacy settings changed to “protected,” or were suspended by Twitter by the time we collected the historical data. Some media reports suggest that Twitter frequently suspends ISIS-related accounts (Berger and Morgan, 2015; Shane and Hubbard, 2014). Specifically, we failed to get the data for nearly 15,000 out of 180,000 users, meaning that our data set is probably biased towards less offensive and less openly hateful users. Despite this shortcoming, our classification results show that even this set of “more politically correct” users could be easily identified.

Needless to say, Twitter users are not representative of the general population. But especially regarding phenomena where there is such strong polarization, surveys would also suffer from non-response bias, as participants might prefer not to openly voice support for a militant organization. In that sense, Twitter with its perceived anonymity lends itself to the study of such phenomena. The fact that our findings concerning the frustration with failed Arab Spring revolutions are in line with previous research (Kohlmann and Alkhouri, 2014; Plous and Zimbardo, 2004) further suggests that our results are not mere artifacts of user selection bias.

Our definition of the “before” period could also be seen as a limitation. As the cut-off date between the “before” and “after” period is different for each user and depends on the date of their first on-topic tweet, it is hard to generalize to the population level. This also relates to the possibility that the temporal dynamics are likely to change over time and that the early supporters might differ from supporters joining later. The fact that our classifier achieves such good performance suggests, though, that there are not too many temporal subgroups with vastly different characteristics.

Sociological interpretation

As our results show that initial support for the Arab Spring often translated into later support for ISIS, it is instructive to situate the dynamics of support in the context of social movements, and to clarify the relationships between them. To this end, we here discuss (i) framing structures, (ii) political opportunity, and (iii) mobilization structures.

Framing structures

Framing structures help movement participants negotiate a common understanding or a framing of causes, remedy, and required action (Benford and Snow, 2000). There seem to be two social movements underlying the observed dynamics, namely the Arab Spring and Islamism, which often intersect. We first provide a brief background on both movements, and then we discuss framing.

Islamism’s central ideology is the desire to make Islam the core underpinning of political, economic, and social constructs, and Islamist social movement organizations (SMOs) have been subjected to frequent crackdowns by Arab regimes (Moghadam, 2013). Islamist SMOs cover a wide spectrum, ranging from those that are strictly political to those that use violence, with ISIS perhaps being the most extreme (Moghadam, 2013). The Arab Spring, in contrast, was started by two non-religious societal components, namely workers demanding economic reform and “urban intelligentsia” demanding democracy (Wulf, et al., 2013).

In terms of framing, both movements identified existing regimes as the main impediment to achieving their goals and the main source of injustice, and both deemed that these regimes must be changed (Haas and Lesch, 2012). Islamists participated widely in the Arab Spring, leading to initial Arab Spring success with the removal of the presidents of Tunisia, Egypt, and Yemen (Hoffman and Jamal; 2012, Wulf, et al., 2013). In other countries, the Arab Spring was met with violent repression (as in Syria) or was circumvented using other means (as in Saudi Arabia). Because the Islamists were well organized, they were able to win elections in Tunisia, Egypt, and Libya, and played an effective role in organizing the opposition in Yemen and Syria (Al-Anani, 2012). Thus, Islamists banded together with other components of the Arab Spring to overthrow the autocratic regimes, although they did not necessarily agree on the desired form of post-Arab Spring regimes, as was evident in Egypt, Tunisia, Syria, and Libya.

Political opportunity

Political opportunity refers to opportunity for change due to: openness of political participation; division between a society’s elite; support of some of the elite for change; and diminished capacity of the regime for repression (Meyer, 2004). Despite the initial successes in overthrowing dictators in some countries, political opportunity waned and anti-Islamism and anti-Arab Spring forces fought back (Borge-Holthoefer, et al., 2015). These forces were able to orchestrate violent military takeovers in countries such as Egypt and Yemen, where political participation was severely restricted and repression increased dramatically (Amnesty International, 2015). The loss of political opportunity is apparent in the top hashtags predicting future support for ISIS. For example from Egypt, the hashtag #elect_the_pimp appeared right before the presidential election, the legitimacy of which was challenged by international observers, and which the field marshal Sisi won by 93.2 percent (Kingsley, 2014); and the hashtag #Rabia_Adawiyya referenced the sit-in that was forcefully disbanded, resulting in at least several hundred deaths. Similar examples emerged from Saudi Arabia, such as #Alburayda_sit-in and #million_man_march_for_Haila_Qusair, indicating marches that were protesting the incarceration of activists.

Mobilization structures

Mobilization structures refer to mechanisms that social movements employ, including: SMOs that act as core rational actors to work towards clear goals, weighing costs and benefits and using resources to achieve measurable outcomes (Jenkins, 1983); and tactical repertoires, which constitute different collective action methods (Savage and Monroy-Hernandez, 2015). The Arab Spring seemed to be failing with the overthrow of post-revolution presidents in Egypt (in July 2013) and Yemen (in January 2015), the transformation of the Syrian revolution into a protracted civil war (starting in 2012), and the emergence of a war lord Libya (in February 2014). However, the desire for change had not ceased, and the proposed remedy from the Islamists and the Arab Spring movements continued to be to change the autocratic regimes. Factions within the movements began to consider violence as a justified means to further their goals (Della Porta, 2008). Although ISIS lies at the fringe of the Islamism movement, our findings suggest that ISIS benefited from this shift in attitudes in favor of employing violence. Some of the top hashtags used by pro-ISIS users show a strong engagement with ISIS's propaganda material and organizations (e.g., #TruthOfIslamicState and #ItisamFoundation) and support for ISIS's animosity against regimes (#IslamicStateToLiberateKSA). Also, an Egyptian ISIS affiliate in the Sinai Peninsula used video footage of violence directed against protesters by Egyptian security forces as a resource to attract recruits [10] (#SinaiState).

Our work suggests that support for ISIS is not necessarily ideological. Despite the divergence between the goals of the Arab Spring and ISIS, many Arab Spring supporters may have bought into the motivational frame of ISIS and started supporting violence after the apparent failure of peaceful means in bringing about successful change. Further, our results show that opposition to ISIS seems to come from other Islamists; this opposition may be grounded in the divergence between the Islamism and Arab Spring movements and the divergence between ISIS and the mainstream Islamism movement.

 

++++++++++

Conclusion

In this paper, we presented a simple yet effective methodology to study antecedents of online polarization: for Twitter users of a currently known ideology, we used historical data to look at their tweets before they first expressed that ideology.

Using tweets about ISIS as a data set, we managed to predict future support of or opposition to ISIS with 87 percent accuracy, using only tweets from the pre-ISIS period. An examination of discriminating hashtags suggested that a major source of support for ISIS stems from frustration with the missteps of the Arab Spring. As for opposition to ISIS, it is linked with support for other rebel groups, mostly in Syria, that have been targeted by ISIS, support for existing Middle Eastern regimes, and Shia sectarianism.

Our methods and findings thus contribute to a more dispassionate discussion of ISIS that moves away from a focus on current atrocities and the urgency of the moment in an attempt to shed light on how this terrorist organization managed to grow so quickly. End of article

 

About the authors

Walid Magdy is a scientist at the Qatar Computing Research Institute. His research interests are in information retrieval, social computing, and data science in general. He received his Ph.D. from Dublin City University in Ireland. Before his Ph.D., he worked at IBM then Microsoft as a research engineer. Walid Magdy has over 50 peer-reviewed publications and nine patents.
E-mail: wmagdy [at] qf [dot] org [dot] qa

Kareem Darwish is a senior scientist at the Qatar Computing Research Institute with interests in information retrieval, digital libraries, and natural language processing. Kareem Darwish consulted with Kevric, a bioinformatics firm in Maryland, and worked as a researcher at the Cairo Microsoft Innovation Lab and the IBM Human Language Technologies group in Cairo. He also taught in the Electrical Engineering Department at the German University in Cairo and the Faculty of Computer and Informatics at Cairo University.
E-mail: kdarwish [at] qf [dot] org [dot] qa

Ingmar Weber is a senior scientist in the Social Computing group at the Qatar Computing Research Institute. His research focuses on using large amounts of online data to study offline phenomena, including political unrest, obesity, gender inequalities, international migration, religion, and relationship breakups. He has published over 80 peer-reviewed articles and has worked with medical doctors, demographic researchers, sociologists, political scientists, and sociolinguists. With Yelena Mejova and Michael Macy he edited Twitter: A digital socioscope (Cambridge: Cambridge University Press, 2015).
E-mail: iweber [at] qf [dot] org [dot] qa

 

Notes

1. http://www.ft.com/cms/s/2/97130d46-7952-11e4-9567-00144feabdc0.html.

2. http://www.newsweek.com/german-journalist-returns-time-isis-chilling-stories-293781.

3. Louis, 2009, p. 125.

4. http://www.reuters.com/article/idUSL1N0RI0R520140917.

5. Twitter lists are manually curated lists of Twitter accounts on a given topic of the curator’s choice. As an example https://twitter.com/JuanFlx/lists/social-media-experts contains a list of accounts experts in social media.

6. Freebase is a taxonomy of topics that YouTube uses to classify its videos. See https://developers.google.com/youtube/v3/guides/searching_by_topic for details.

7. A roll call vote guarantees that every Member’s vote is recorded, but only a minority of bills receive a roll call vote; http://thomas.loc.gov/home/rollcallvotes.html.

8. Using the language operator “lang:ar” on Twitter’s streaming API would have resulted in a dramatically reduced data set, as the operator is provided on a post-sample basis, i.e., after sampling down to one percent of tweets. Therefore we used the Twitter’s search API instead.

9. The small loss in accuracy was due to five accounts that were automatically labeled as “anti-ISIS” but were deemed “neutral” by one of the annotators. Two of these five accounts were seen as “anti-ISIS” by the second annotator.

10. https://www.youtube.com/watch?v=hi6esvr06cQ.

 

References

R.A. Abdulla, 2007. “Islam, jihad, and terrorism in post-9/11 arabic discussion boards,” Journal of Computer-Mediated Communication, volume 12, number 3, pp. 1,063–1,081.
doi: http://dx.doi.org/10.1111/j.1083-6101.2007.00363.x, accessed 17 January 2016.

L.A. Adamic and N. Glance, 2005. “The political blogosphere and the 2004 U.S. election: Divided they blog,” LinkKDD ’05: Proceedings of the Third International Workshop on Link Discovery, pp. 36–43.
doi: http://dx.doi.org/10.1145/1134271.1134277, accessed 17 January 2016.

W.H. Allendorfer and S.C. Herring, 2015. “ISIS vs. the U.S. government: A war of online video propaganda,” First Monday, volume 20, number 12, at http://firstmonday.org/article/view/6336/5165, accessed 17 January 2016.
doi: http://dx.doi.org/10.5210/fm.v20i12.6336, accessed 17 January 2016.

A.J. Bellamy, 2009. “Torture, terrorism, and the moral prohibition on killing non-combatants,” In: W.G.K. Stritzke, S. Lewandowsky, D. Denemark, J. Clare, and F. Morgan (editors). Terrorism and torture: An interdisciplinary perspective. Cambridge: Cambridge University Press, pp. 18–43.
doi: http://dx.doi.org/10.1017/CBO9780511581199.003, accessed 17 January 2016.

R.D. Benford and D.A. Snow, 2000. “Framing processes and social movements: An overview and assessment,” Annual Review of Sociology, volume 26, pp. 611–639.
doi: http://dx.doi.org/10.1146/annurev.soc.26.1.611, accessed 17 January 2016.

J.M. Berger and J. Morgan, 2015. “The ISIS Twitter census: Defining and describing the population of ISIS supporters on Twitter,” Brookings Project on U.S. Relations with the Islamic World, Analysis Paper, number 20, at http://www.brookings.edu/~/media/research/files/papers/2015/03/isis-twitter-census-berger-morgan/isis_twitter_census_berger_morgan.pdf, accessed 17 January 2016.

J. Borge-Holthoefer, W. Magdy, K. Darwish, and I. Weber, 2015. “Content and network dynamics behind Egyptian political polarization on Twitter,” CSCW ’15: Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work & Social Computing, pp. 700–711.
doi: http://dx.doi.org/10.1145/2675133.2675163, accessed 17 January 2016.

C. Budak and D.J. Watts, 2015. “Dissecting the spirit of Gezi: Influence vs. selection in the Occupy Gezi movement” Sociological Science, volume 2, at https://www.sociologicalscience.com/articles-v2-18-370/, accessed 17 January 2016.
doi: http://dx.doi.org/10.15195/v2.a18, accessed 17 January 2016.

R. Cohen-Almagor, 2012. “In Internet’s way: Radical, terrorist islamists on the free highway,” International Journal on Cyber Warfare and Terrorism, volume 2, number 3, pp. 39–58.
doi: http://dx.doi.org/10.4018/ijcwt.2012070104, accessed 17 January 2016.

M.D. Conover, J. Ratkiewicz, M. Francisco, B. Gon¸alves, F. Menczer, and A. Flammini, 2011. “Political polarization on Twitter,” Proceedings of the Fifth International AAAI Conference on Weblogs and Social Media, at http://www.aaai.org/ocs/index.php/ICWSM/ICWSM11/paper/view/2847, accessed 17 January 2016.

K. Darwish, W. Magdy, and A. Mourad, 2012. “Language processing for Arabic microblog retrieval,” CIKM ’12: Proceedings of the 21st ACM International Conference on Information and Knowledge Management, pp. 2,427—2,430.
doi: http://dx.doi.org/10.1145/2396761.2398658, accessed 17 January 2016.

L. Dearden, 2014. “Isis vs Islamic State vs Isil vs Daesh: What do the different names mean — and why does it matter?” Independent (23 September), at http://www.independent.co.uk/news/world/middle-east/isis-vs-islamic-state-vs-isil-vs-daesh-what-do-the-different-names-mean-9750629.html, accessed 17 January 2016.

D. Della Porta, 2008. “Research on social movements and political violence,” Qualitative Sociology, volume 31, number 3, pp. 221–230.
doi: http://dx.doi.org/10.1007/s11133-008-9109-x, accessed 17 January 2016.

J.P. Farwell, 2014. “The media strategy of ISIS,” Survival, volume 56, number 6, pp. 49–55.
doi: http://dx.doi.org/10.1080/00396338.2014.985436, accessed 17 January 2016.

M.L. Haas and D.W. Lesch (editors), 2012. The Arab Spring: Change and resistance in the Middle East. Boulder, Colo.: Westview Press.

J. Horgan, 2003. “The search for the terrorist personality,” In: A. Silke (editor). Terrorists, victims, and society: Psychological perspectives on terrorism and its consequences. Chichester: Wiley, pp. 3–27.

A.A. Jamal, R.O. Keohane, D. Romney, and D. Tingley, 2015. “Anti-Americanism and anti-interventionism in Arabic Twitter discourses,” Perspectives on Politics, volume 13, number 1, pp. 55–73.
doi: http://dx.doi.org/10.1017/S1537592714003132, accessed 17 January 2016.

J.C. Jenkins, 1983. “Resource mobilization theory and the study of social movements,” Annual Review of Sociology, volume 9, pp. 527–553.
doi: http://dx.doi.org/10.1146/annurev.so.09.080183.002523, accessed 17 January 2016.

J. Klausen, 2015. “Tweeting the jihad: Social media networks of western foreign fighters in Syria and Iraq,” Studies in Conflict & Terrorism, volume 38, number 1, pp. 1–22.
doi: http://dx.doi.org/10.1080/1057610X.2014.974948, accessed 17 January 2016.

E. Kohlmann and L. Alkhouri, 2014. “Profiles of foreign fighters in Syria and Iraq,” CTC Sentinel (29 September), at https://www.ctc.usma.edu/posts/profiles-of-foreign-fighters-in-syria-and-iraq, accessed 17 January 2016.

Z. Liu and I. Weber, 2014. “Is Twitter a public sphere for online conflicts? A cross-ideological and cross-hierarchical look,” In: L.M. Aiello and D. McFarland (editors). Social Informatics. Lecture Notes in Computer Science, volume 8851. Berlin: Springer, pp. 336–347.
doi: http://dx.doi.org/10.1007/978-3-319-13734-6_25, accessed 17 January 2016.

W.R. Louis, 2009. “If they’re not crazy, then what? The implications of social psychological approaches to terrorism for conflict management,” In: W.G.K. Stritzke, S. Lewandowsky, D. Denemark, J. Clare, and F. Morgan (editors). Terrorism and torture: An interdisciplinary perspective. Cambridge: Cambridge University Press, pp. 125–153.
doi: http://dx.doi.org/10.1017/CBO9780511581199.008, accessed 17 January 2016.

W. Magdy, K. Darwish, and I. Weber, 2015. “‘I like ISIS, but I want to watch Chris Nolan’s new movie’: Exploring ISIS supporters on Twitter,” HT ’15: Proceedings of the 26th ACM Conference on Hypertext & Social Media, pp. 321–322.
doi: http://dx.doi.org/10.1145/2700171.2794352, accessed 17 January 2016.

D.S. Meyer, 2004. “Protest and political opportunities,” Annual Review of Sociology, volume 30, pp. 125–145.
doi: http://dx.doi.org/10.1146/annurev.soc.30.012703.110545, accessed 17 January 2016.

L. Miller, 2006. “The terrorist mind: I. A psychological and political analysis,” International Journal of Offender Therapy and Comparative Criminology, volume 50, number 2, pp. 121–138.
doi: http://dx.doi.org/10.1177/0306624X05281405, accessed 17 January 2016.

V.M. Moghadam, 2013. Globalization and social movements: Islamism, feminism, and the global justice movement. Second edition. Lanham, Md.: Rowman & Littlefield.

F. Morstatter, J. Pfeffer, H. Liu, and K.M. Carley, 2013. “Is the sample good enough? Comparing data from Twitter’s streaming API with Twitter’s firehose,” Proceedings of the Seventh International Conference on Weblogs and Social Media, pp. 400–408, and at http://www.aaai.org/ocs/index.php/ICWSM/ICWSM13/paper/view/6071, accessed 17 January 2016.

D. O’Callaghan, N. Prucha, D. Greene, M. Conway, J. Carthy, and P. Cunningham, 2014. “Online social media in the Syria conflict: Encompassing the extremes and the in-betweens,” 2014 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM), pp. 409–416.
doi: http://dx.doi.org/10.1109/ASONAM.2014.6921619, accessed 17 January 2016.

S.L. Plous and P.G. Zimbardo, 2004. “How social science can reduce terrorism,” Chronicle of Higher Education (10 September), pp. B9–B10.

K.T. Poole and H. Rosenthal, 1985. “A spatial model for legislative roll call analysis,” American Journal of Political Science, volume 29, number 2, pp. 357–384.
doi: http://dx.doi.org/10.2307/2111172, accessed 17 January 2016.

P. Ross, 2014. “ISIL, ISIS, Islamic State, Daesh: What’s the difference?” International Business Times (23 September), at http://www.ibtimes.com/isil-isis-islamic-state-daesh-whats-difference-1693495, accessed 17 January 2016.

R. Sanchez, 2015. “ISIS, ISIL or the Islamic State?” CNN (23 January), at http://www.cnn.com/2014/09/09/world/meast/isis-isil-islamic-state/, accessed 17 January 2016.

S. Sarangi and L. Alison, 2005. “Life story accounts of left wing terrorists in India,” Journal of Investigative Psychology and Offender Profiling, volume 2, number 2, pp. 69–86.
doi: http://dx.doi.org/10.1002/jip.30, accessed 17 January 2016.

S. Savage and A. Monroy-Hernández, 2015. “Participatory militias: An analysis of an armed movement’s online audience,” CSCW ’15: Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work & Social Computing, pp. 724–733.
doi: http://dx.doi.org/10.1145/2675133.2675295, accessed 17 January 2016.

S. Shane and B. Hubbard, 2014. “ISIS displaying a deft command of varied media,” New York Times (30 August), at http://www.nytimes.com/2014/08/31/world/middleeast/isis-displaying-a-deft-command-of-varied-media.html, accessed 17 January 2016.

Y. Tyshchuk, W.A. Wallace, H. Li, H. Ji, and S.E. Kase, 2014. “The nature of communications and emerging communities on Twitter following the 2013 Syria sarin gas attacks,” 2014 IEEE Joint Intelligence and Security Informatics Conference (JISIC), pp. 41–47.
doi: http://dx.doi.org/10.1109/JISIC.2014.16, accessed 17 January 2016.

I. Weber, V.R. Kiran Garimella, and A. Batayneh, 2013. “Secular vs. Islamist polarization in Egypt on Twitter,” ASONAM ’13: Proceedings of the 2013 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining, pp. 290–297.
doi: http://dx.doi.org/10.1145/2492517.2492557, accessed 17 January 2016.

W. Weerkamp, S. Carter, and M. Tsagkias, 2011. “How people use Twitter in different languages,” Proceedings of the ACM WebSci’11, pp. 1–2, and at http://www.websci11.org/fileadmin/websci/Posters/90_paper.pdf, accessed 17 January 2016.

 


Editorial history

Received 24 December 2015; revised 14 January 2016; accepted 17 January 2016.


Creative Commons License
This paper is in the Public Domain.

#FailedRevolutions: Using Twitter to study the antecedents of ISIS support
by Walid Magdy, Kareem Darwish, and Ingmar Weber.
First Monday, Volume 21, Number 2 - 1 February 2016
http://www.firstmonday.org/ojs/index.php/fm/article/view/6372/5194
doi: http://dx.doi.org/10.5210/fm.v21i2.6372





A Great Cities Initiative of the University of Illinois at Chicago University Library.

© First Monday, 1995-2017. ISSN 1396-0466.