A Qualitative and Quantitative Method for Studying Religious Virtual Communities: The Case of the Salafi United Kingdom’s Community on Twitter (X)

Abstract

This open-source-based article presents an automated method for identifying and tracing popular Salafi discussions online. The novelty of this method lies in its inter-disciplinary approach developed through collaboration among experts in the fields of the Middle East, Islamic studies, and computer science. The computerized model presented here harnesses machine learning techniques to accurately identify popular Salafi writings on social media and to distinguish them from the writings of Muslims from other denominations. Creating an AI-supported model to distinguish between writings on social media that pertain to two different Islamic denominations is a highly difficult task. Based on this machine learning model and the methodology that it implements, the study presented here identifies United Kingdom-based Twitter accounts that embody Salafi thinking (even if they do not utilize terminology that is manifestly Salafi) and, based on that identification, analyzes and characterizes the United Kingdom-based Salafi community on Twitter. Unlike other machine learning ideology-related studies that are focused on Salafi-jihadism, the present research is focused on quietist Salafism (Salafi-taqlidis) in the United Kingdom. The purpose of this study is to examine the virtual Salafi community in the United Kingdom, with a focus on identifying the key issues of concern to its members and assessing the influence of global Salafi trends within this UK-based community.

1. Introduction

Prior to the internet era, religious communities and religious knowledge evolved through physical interactions. Religious centers, religious seminaries (e.g., madrasa, beit midrash), houses of worship, and public squares served as hubs where people could share knowledge and ideas through lectures, printed material, recordings, and other face-to-face communications with co-religionists. The advent of the World Wide Web in the early 1990s and social media in the early 2000s introduced a new channel through which religious communities and religious knowledge could evolve. People could now access information through religiously oriented platforms such as discussion forums, digital libraries, online lectures, and sites maintained by religious scholars and/or institutions and could interact with co-religionists through social media. Entire communities could exist without people ever having to meet face-to-face. Virtual religious communities no longer needed physical buildings, meeting rooms, or places of worship to exist. Furthermore, because the internet was an impartial platform where no single person or entity, including religious scholars, could dominate the discussion, it inevitably extended the spectrum of opinions and perspectives that became part of the legitimate discourse within a given religion. Unlike in traditional religious institutions, where mostly erudite scholars could share their ideas, in cyberspace, all people may partake in the religious discussion regardless of the level of their knowledge. Accordingly, the entire concept of what religious knowledge is, and who possessed such knowledge, changed drastically.

The emergence of online religious knowledge and virtual religious communities poses a challenge to scholars undertaking to study modern religions. Some of the questions that vex contemporary students of, for example, modern Islam include the following: How can one measure the expansion of Islamic communities when believers do not necessarily belong to any physical community? How can one determine whether an online community exists as such? What constitutes a virtual religious community? Is there a methodology for determining which social media accounts participating in an online discussion are truly Islamic? The dilemma is aggravated when attempting to distinguish between writings representing two different denominations within Islam. For example, whereas most experts on Islam can easily differentiate between Salafi and non-Salafi Islamic scholarly writing, they are often confused when faced with the unsystematic and often casual writings by Salafis and non-Salafi Muslims on social media. Distinguishing between Salafi and non-Salafi, albeit Islamic, social media accounts is, many times, a difficult task, as demonstrated below.

This study, which is part of a growing field known as digital humanities (Carter 2013; Romanov 2018; Muhanna 2016),1 advanced an automated method for identifying and tracing Salafi discourse online. Through collaboration between experts in the fields of Middle Eastern studies, Islamic studies, Salafism, and computer science, the authors developed highly accurate computerized models based on machine learning techniques, which can identify Salafi writings on social media and distinguish them from writings by Muslims of other denominations. Using AI-supported models, this project first automatically identifies Twitter accounts based in the United Kingdom that reflect Salafi thought. Second, it maps the geographical distribution of this virtual community, highlighting the UK cities where Salafi Twitter users are most concentrated. Third, by analyzing thousands of tweets, the project offers preliminary insights into current Salafi trends within the UK’s virtual Salafi community. In future work, the authors will employ this automated method to more comprehensively delineate existing Salafi trends within the UK’s Salafi virtual community. Additionally, the authors will analyze Twitter-based discussions among Salafis on critical issues such as social integration into broader UK society, participation in parliamentary elections, engagement with state authorities and institutions (e.g., holding governmental positions or using state courts), and the legitimacy of proclaiming takfir against ‘deviant’ Muslims (e.g., Muslim Brothers and Sufis). By continuously analyzing these discussions, the authors, in future studies, aim to trace the evolution of religious thought within the UK Salafi virtual community.

2. The Current State of Research

Various fields utilize machine learning today. These include, among others, cyber security, healthcare, terrorism, intelligent transportation systems, smart grids, and computer networks (Jhaveri et al. 2022) for functionalities such as detection, prediction, analysis, and identification (Angra and Ahuja 2017). Until now, machine learning has been used in Islamic studies primarily for detection and prediction. For example, experts in computer science used machine learning for detection in the fields of Hadith studies, Quranic studies, and Islamic law studies to determine whether a hadith is authenticated (Sulistio et al. 2024; Abdelaal and Youness 2019). They created a chatbot, also known as a virtual assistant, that Muslims could use to answer questions and interpret terms in Islamic law (Jamal et al. 2020). Machine learning has also proven useful in finding a correlation between the teachings of the Quran and the teachings of the Hadith (Rostam and Malim 2021).

Computer science experts have also employed machine learning methods for detection in the field of radical Islamic communication. Most of these projects focused on tracing linguistic features typically found in communications by terrorist groups, such as ISIS and al-Qaeda, to facilitate automatic detection and a quick shutdown of social media accounts whose content supports terrorism. Some of the studies adopted various approaches to semantic analysis, such as hashtag analysis (Ashcroft et al. 2015; Yang et al. 2011; Magdy et al. 2015), term-based approach (Rowe and Saif 2016), emotion lexicon (Abbasi and Chen 2007), and classifiers to examine affect (Chen 2008) in order to assess whether the writer is pro-ISIS or anti-ISIS. Other studies analyzed the frequency of context-dependent and context-independent linguistic features (Kaati et al. 2015). A unique study employed a method called Reciprocal Human Machine Learning to allow better detection of radical accounts (Lewinsky et al. 2024). Machine learning was also applied for prediction in the context of health-related research pertaining to Muslim societies. For example, scholars used it to identify potential risk factors affecting malnourished women in Bangladesh (Islam et al. 2022a), to predict childhood anemia among children in Bangladesh (Khan et al. 2019), and to predict breast cancer among women in Bangladesh (Islam et al. 2022b).

As religious virtual communities proliferate in cyberspace, scholars of modern Islam are forced to employ machine learning methods in order to identify, study, and analyze such communities. This study attempts to solve one of the primary challenges that face contemporary scholars of Salafism, namely, identifying and assessing the evolution in size and in thought of Salafi virtual communities. The machine learning-based study presented here is unique in two ways: First, as explained more elaborately in a separate section below, it does not rely solely on detecting linguistic features such as keywords, prominent terms, and names but rather operates by teaching an algorithm to identify some of the fundamentals of the Salafi ideology regardless of how they are expressed by a specific writer. In addition, unlike most machine learning ideology-related studies, the model presented here focuses on quietist Salafism (Salafi-taqlidis) in the United Kingdom rather than militant Salafism (Salafi-jihadis).

3. Contemporary Salafism

Salafis are Muslims who maintain that believers must live their lives in the same way that the Prophet Muhammad lived his. They strictly and exclusively adhere to the Qur’an, the Hadith (i.e., assumed records of the Prophet’s conduct), and the interpretation of these sources promulgated by the first three generations of Muslims (al-salaf al-salih) (Lauziere 2016, p. 201; Haykel 2009, p. 34; Maher 2016, p. 7).2 Salafis embrace a strict concept of tawhid (the oneness of God), which consists of three indivisible tenets: God is the sole creator and sovereign of the universe (tawhid al-rububiyya), God alone has the right to be worshiped (tawhid al-uluhiyya), and God’s characteristics and powers are unique, not manifested in any person or entity (tawhid al-asmaʼ wal-sifat) (Wiktorowicz 2006, p. 209).

In a seminal article published in 2006, Quintan Wiktorowicz classifies Salafis into three distinct groups: Purists (also known as taqlidis, traditional Salafis or quietists), Politicos, and Jihadis. He claims that all three groups adhere to the same doctrine (ʻaqida) and strive to create regimes that are purely Muslim, but they disagree on the proper way to implement the doctrine in modern times (manhaj) (Wiktorowicz 2006, p. 221; Alshech 2014). The Purists, consisting primarily of older generations of Salafi scholars who dominate the Saudi religious establishment, eschew politics and political activism and focus instead on purifying Islam through preaching and education (tasfiyya wa-tarbiyya). Their goal is to purify Islam of religious innovations and restore it to the original form practiced by the first three Muslim generations. According to the Purists, political and military activism would only exacerbate the moral chaos already present in contemporary Muslim societies. The Politicos, who follow a younger generation of scholars originating in the Saudi kingdom, challenged the Purists’ authority in the 1980s and 1990s and claimed, inter alia, that citizens are required to exhort their governing regimes, both privately and publicly, to promote a perfect Islamic state. Influenced by the dissident members of the Muslim Brotherhood who fled Egypt and sought refuge in Saudi Arabia, the Politicos believed that achieving moral change requires political involvement, public activism, and demonstrations (Lacroix 2011, p. 51). The third group, the Jihadis, appeared in the modern Saudi political arena in the mid-1990s (Moghadam 2008, p. 14). They regarded both quietism and political engagement as ineffective methods for addressing what they perceived as the Islamic nation’s critical moral decline. Instead, they sought to bring about immediate change through relentless jihad against both Muslim and non-Muslim governments (Alshech 2014, p. 159; Hatina 2021, p. 98).3 Though both the Politicos and the Jihadi Salafis opposed the Purists, they differed considerably in their ideological orientations. The Jihadis sanctioned the use of violence and the proclamation of takfir (i.e., an accusation of heresy against Muslims) against Muslim rulers, while the Politicos explicitly denounced these methods (Azani and Koblentz-Stenzler 2019).4

For years, Wiktotowicz’s typology was widely accepted by scholars, journalists, and policy makers, serving as a fundamental framework for analyzing Salafism (Amghar and Cavatorta 2023, p. 196; Gauvain 2013, p. 12; Sinani 2022). Some scholars have sought to refine these categories to account for local political shifts, particularly after the Arab Spring (Wagemakers 2016, p. 17).5 Recently, however, Wiktorowicz’s categorization has faced significant criticism on several fronts. First, scholars have questioned his assumption that Jihadis and Purists share the same doctrine, arguing instead that they represent two distinct and unrelated social categories (Amghar 2023, p. 207).6 Second, they have criticized the rigidity of Wiktorowicz’s categories, claiming that they often fail to capture the complexity of Salafi views. For instance, the Purists’ opposition to jihad is not intrinsic but conditional: Purists may support defensive jihad against a foreign invader while rejecting it against Muslim rulers, especially within their own country (Blanc 2023, p. 4). Moreover, as Weismann indicates, in some locations, “Salafi thinkers themselves had roots in the revivalist Sufi tradition of the previous centuries” despite the common understanding among academic scholars that Salafis abhor Sufi practices, labeling some of them as shirk (Weismann 2005, p. 39). Third, scholars have pointed out the ambiguity of Wiktorowicz’s categories. Quietism, for example, is not necessarily apolitical but can be a deliberate withdrawal from politics, as expressed by Shaykh al-Albani (d. 1999) in his statement, “part of politics is leaving the politics (Bonnefoy 2016)”. Additionally, Purists may engage in political activities without formally participating in party politics (Deschamps-Laporte 2023).7 Fourth, critics have argued that Wiktorowicz’s typology is overly rigid, thus failing to account for evolving attitudes among Salafis in the post-revolutionary MENA region. For instance, Purists established political parties in Egypt, Tunisia, and Yemen, while Purist loyalists, notably the Madkhalis, formed armed militias in Libya (Blanc 2023, p. 9; Karagiannis 2019, p. 207; Deschamps-Laporte 2023, p. 232).8 Finally, scholars contend that Wiktorowicz’s typology is too culturally specific, primarily reflecting the contexts of Jordan and Saudi Arabia, and does not adequately account for the ideological and political diversity of Salafism in regions such as Europe, Asia, and beyond (Adraoui 2023, p. 289; Pall 2023, p. 271).9

It is important to note, however, that, in cases where Salafi communities developed through sustained engagement with Saudi Arabia—most notably when local students received Saudi-funded religious education and subsequently returned to teach in their home countries—local Salafi scholars often aligned with either Purist or Politico currents. A notable example is the emergence of the Salafi community in the United Kingdom (Hamid 2009, p. 394).

4. The Emergence of Salafism in the United Kingdom

Salafism took roots in the United Kingdom as early as the 1960s when shaykh Fazal Karim Asim, who migrated to the United Kingdom from Pakistan in 1962, commenced his da‘wa (i.e., missionary) activities in the mosque at the United Kingdom Islamic Mission (UKIM) before joining the Ahl al-Hadith Mosque in the UK. At the time, the Ahl al-Hadith community (i.e., people who refuse to follow legal opinions not based on Qur’an and Hadith) was small and relatively unknown, partially because it lacked English-speaking scholars and preachers (Dawood 2021, p. 78). In the 1970s, however, Saudi Arabia accelerated its da‘wa activities within the United Kingdom. A da‘wa delegation from the Islamic University of Medina, including shaykh al-Albani, was sent to strengthen the Ahl al-Hadith da‘wa in Birmingham in 1975. “The Ahl al-Hadith was thus a key ally of Saudi Arabia and with the help and financial assistance of Saudi Arabia, played an important role in introducing Salafism as… a new moral practical approach to life which could potentially become hegemonic… in the UK” (Dawood 2021, p. 78). The Saudi state also granted scholarships for British students to study at the Islamic University of Medina, hundreds of whom later returned to teach and spread the Salafi creed in the UK (Brit 2004, p. 171).

The Saudi quest to propagate Salafism internationally in order to gain legitimacy and dominance among other Islamic ideologies (Dawood 2021, p. 38) was successful partially due to an abundance of petro-dollars. Saudi Arabia was able to repulse pan-Arabism and Shi’ism (particularly following the Islamic revolution in Iran in 1979) on a global scale by establishing international organizations such as the Muslim World League and the World Assembly of Muslim Youth, which founded branches in various European countries including the United Kingdom (Al-Rasheed 2005, p. 155).10 The Saudi government used these international institutions to publish and distribute Salafi literature in English and other European languages. Saudi Arabia also funded mosques in the United Kingdom in cities such as London, Edinburgh, Leicester and Birmingham and established other Islamic institutions, all with the goal of expanding the influence of Salafi ideology (Al-Rasheed 2005, p. 151).11 Alongside Saudi Arabia, Salafi institutions based in Kuwait, Yemen, Jordan, Egypt, and South Asia have all played a role in the dissemination of Salafism in the United Kingdom (Dawood 2021, p. 40).

The second generation of Salafis in the United Kingdom, some of whom are alumni of the Islamic University of Medina, established the Jami’yyat Ihyaa Manhaj al-Sunnah (JIMAS) (The Association for Reviving the Prophet’s Methodology) in 1984. JIMAS “functioned as an umbrella organization coordinating a national movement” (Dawood 2021, p. 79). At the outset, most JIMAS leaders and activists were of Asian background (e.g., Abu Muntasir, Abu Aliyah, Uzma, and Haniya) (Hamid 2009, p. 387), but this changed in the 2000s when Shaykh Haitham al-Haddad, of Palestinian origin, emerged on the United Kingdom Salafi scene (Dawood 2021, p. 80). Soon, JIMAS linked with Salafi scholars and speakers worldwide and was connected to the global Salafi community in the Middle East, South Asia, the United States, and Australia. Salafi youth in East London undertook to spread Salafi ideology among the larger Islamic community and to seek out new members. Their da‘wa efforts to convert people to Salafism were not only aimed at Muslims but were also directed at British society as a whole. British people who were attracted to the Salafi doctrine were mainly second-generation South Asian Muslims aged 18–30 with some additional significant followings of black and white converts.

By the mid-1990s, additional Salafi centers were established throughout the United Kingdom and, despite their connection to JIMAS, they displayed their own independent thinking. At that time, friction between Salafis and other British Muslim denominations grew. Non-Salafi Muslims disagreed with Salafis’ longstanding rejection of traditional religious understandings shaped by Medieval Islamic scholarly interpretations. Salafis viewed such traditional conceptions as taqlid, i.e., the forbidden blind following of human authorities rather than the divine authorities (Hamid 2009, p. 390).

The friction between the Salafi Purists, Politicos, and Jihadis in Saudi Arabia, which resulted in part from Saudi Arabia’s permitting the United States to build permanent military bases in the kingdom in 1991, had an impact on the Salafi community in the United Kingdom. “While JIMAS and individuals associated with them had made links with the Politicos in Saudi Arabia who formed the Sahwa movement, which opposed the Saudi government’s decision to invite the Americans into the Kingdom, a pro Saudi government member, Abdul Wahid (also known as Abu Khadeejah), led the opposition within JIMAS and supported the Puritans in Saudi Arabia who approved the regime’s decision (Brit 2004, p. 172).12 He challenged Abu Muntasir and those who were sympathetic to the Sahwa movement, resulting in the fragmentation of the organization over the period of a year” (Hamid 2009, p. 394). This led to an internal break and the creation of a splinter group in the United Kingdom: The Organization of Ahl al-Sunnah Islamic Societies (OASIS).13 Around the time that the Purists and Politicos were quarreling, a third trend, Salafi-jihadism, was emerging in the UK, led by the Jamaican convert Abdullah al-Faisal and the Jordanian/Palestinian Abu Qatada al-Filastini (Hamid 2009, p. 396).

Thus, by the end of the 1990s, there were three primary ideological trends among British Salafis. Contemporary academic scholars who study UK-based Salafism are faced with the challenge of monitoring the spread of the three distinct Salafi movements. This challenge is especially daunting in the context of online Salafi communities whose debates and disagreements are primarily expressed on social media. To date, no methodology has been proposed to effectively accomplish this task.

5. The Unique Features of Our AI-Supported Model

Salafi discourse on social media often lacks the explicit lexical markers or ideological keywords that scholars typically use to identify a writer’s ideological stance. Without these linguistic indicators, informal Salafi communications on social media may resemble those of non-denominational Muslims. For instance, consider the following Salafi tweet: “Allah judges with truth while those whom they invoke besides Him cannot judge anything”. Although this statement can be interpreted as a general Islamic affirmation of Allah’s role as the ultimate judge, Salafi experts would recognize it as an expression of tawhid al-asma’ wa’l-sifat—the Salafi belief that God’s attributes and powers are singular and cannot be attributed to any other entity. This belief dictates that, since one of Allah’s attributes in the Qur’an is the Judge (al-hakam), all judgments must be grounded exclusively in His law. Thus, while a lingual approach to this tweet would not have flagged it as Salafi, our trained model was able to do so. Training an AI model to identify core Salafi ideologies in texts that lack explicit Salafi terminology is a distinct challenge, which we addressed in our AI-supported model. We therefore propose two hypotheses:

H1. 

Machine learning models can be designed to effectively identify Salafi discourse on social media and to differentiate it from non-Salafi Muslim discourse.

H2. 

An automated analysis of Twitter discussions can provide insights into the evolving religious and social trends within the UK’s Salafi community.

We chose the United Kingdom as a test case for two reasons. First, Salafi content in the UK is written in English, which made it easier for us to train analysts. Second, given the recent rise in Muslim political activism in the UK, among other Western countries, studying shifts in Salafi sentiments is crucial to contemporary researchers. We chose Twitter as our research platform, despite a larger proportion of British users being on Facebook (around 65%—44 million out of 67 million) compared to Twitter (approximately 30%—23 million out of 67 million), because Twitter’s API is more researcher friendly than Facebook’s. Additionally, we opted not to use TikTok, even though its user base in the UK is similar in size to Twitter’s, because TikTok primarily features short videos, whereas Twitter also includes short text posts, which are crucial for our analysis.

6. Methodology

Data Collection and Labeling

To address the challenge of training an algorithm to recognize core principles of the Salafi creed, rather than merely identifying Salafi-specific keywords, this study employed an alternative approach. A group of 15 analysts were trained by an expert in Salafi ideology on the fundamentals of the Salafi creed.14 The analysts were taught the Salafi ideology for a few weeks and were trained to discern its various manifestations in social media writings. The analysts, as a group, were then asked to identify a total of 25 British Salafi Twitter accounts (in English) and to collect manually dozens of tweets whose content reflect Salafi ideas and tenets. The accounts were scrutinized by a Salafi expert to certify their Salafi affiliation.

Once the analysts were proven capable of correctly identifying Salafi accounts, each analyst was requested to locate 3 British Salafi Twitter accounts in English. A total of 45 Salafi accounts were passed on to experts in computer science who (using Twitter API) extracted thousands of tweets from them. This cluster of tweets was mixed with tweets stemming from randomly picked non-Salafi Muslim users in the UK (the ratio was 1:1 Salafi and non-Salafi), and the entire consolidated set of tweets was then given to the analysts, who were asked to label them as either positive (Salafi), negative (non-Salafi), irrelevant, or unsure based on keywords and text essence. For the sake of efficacy, a unique website was designed where each analyst labeled a total of 980 tweets (for a grand total of 14,700 tweets that were classified). To ensure accuracy of the tweets’ labeling, each tweet was classified by two different analysts, and only tweets that were labeled the same way by both analysts were included in the Salafi dataset. Tweets that received conflicting labeling and “unsure” labeling were transferred to an expert account and were labeled by an expert in Salafi ideology. The tweets that were “irrelevant” were then removed from the dataset. These tagged tweets were used to build a classification model that can automatically identify Salafi tweets online.

Due to privacy concerns, the system only accesses accounts that are open to the public. To safeguard privacy protection for Salafi users online, the data collected do not display the names of the accounts’ owners. The project’s goal is to study the growth and geographical distribution of the Salafi virtual community in the United Kingdom and to trace the various Salafi trends within it. The focus here is the community and its collective views, not any specific Salafi user (For an elaborate description of the Process of Creating the Models, see Appendix A).

7. Results

7.1. Classification Model

A total of 14,700 tweets were collected from 90 twitter UK users (45 Salafi and 45 Non-Salafi Muslim accounts). In total, 3946 tweets were relevant, 1023 labeled as Salafis and 2923 labeled as non-Salafi. These tweets were used to build a model that can automatically identify Salafi tweets online. The most efficient model created (LSTM) displayed 87 percent accuracy (for an elaborate description of the system’s efficiency and accuracy, see Appendix B).

7.2. Salafi Virtual Communities in the UK

As of the time of this writing, the system detected 185 Salafi accounts in the UK, 90 percent of which specified their city of residence. The cities with most Salafi users are London and Birmingham. Other UK cities hosting Salafis are Bradford, Slough, Liverpool, Sheffield, and Leeds.

7.3. Model Is Able to Detect Salafi Content

In the following discussion, we provide some examples indicating that our models already display capability to detect Salafi content even if a tweet lacks most known Salafi keywords. At this point, the models based their successful classification of the tweets as Salafi on the inclusion in those tweets of names of local and international Salafi scholars (e.g., Shaykh Ibn al-‘Uthaymin from Saudi Arabia, Shaykhs Abdul Rahman Hilal and Abu Khadeejah Abdul Wahid from the United Kingdom), the names of Salafi mosques and institutions (mostly in the UK, e.g., Masjid as-Sunnah, Masjid Abdul Aziz Bin Baz), and Salafi keywords (e.g., Salafi aqida). Interestingly, however, some of the tweets that were correctly classified by, for example, the LSTM model, as Salafi did not include any of the above terms or names but instead included content that reflected Salafi thinking only.

Consider the following tweet: “The Pagan Roots of New Year’s Resolutions”, which the LSTM model correctly identified as Salafi, despite the fact that the tweet lacks known Salafi terminology. Salafis of all denominations prohibit celebrating New Year (on 1 January) because they view such a celebration as a forbidden innovation. Salafis endorse and celebrate only two Muslim holidays (‘id al-fitr and ‘id al-adha), which were established at the time of the Prophet Muhammad. In addition, they perceive the celebration of the New Year in January as rooted in Roman pagan culture (circa 46 B.C. emperor Julius Caesar made 1 January the beginning of his new calendar) (Pruitt 2013) and the New Year’s resolution as a secular practice of later times (Al-Akhdar 2018). As a result, Salafis prohibit both the celebration of New Year and the practice of New Year’s resolutions.

Another tweet picked up by the model based on Salafi ideological content rather than keyworks is the following: “If only Ahlul Bid‘ah [i.e., the people of innovation] would accept the truth and admit their mistakes, they wouldn’t be in situations like this. But unfortunately, that’s the ‘reality.’ The Messenger said: Whoever innovates or accommodates an innovator then upon him is the curse of Allah”. Salafis accuse Muslim groups such as the Sufis and the Muslim Brothers of adopting practices (e.g., pilgrimage to Saint graves, participation in parliamentary elections, respectively) that are rooted neither in the Qur’an nor in the Prophetic tradition and label them derogatorily as “the people of innovation”. The concept that Sufis and Muslim Brothers are innovators, which was included in the tweet above, is manifestly Salafi, and the LSTM model correctly identified it as such. Another tweet identified by the LSTM model as Salafi also included content that evidenced the criticism of innovative approaches to Islam: “Whilst Shamsi returns the people back to the Quran and the Sunnah”. Here, the tweet referenced local Salafi preachers in Hyde Park (in Speakers’ Corner) trying to convert people to Salafi ideology. The notion that Salafism epitomizes the Qur’an and Sunna, and that innovative approaches do not, is manifestly Salafi.

Another flagged tweet alludes to the Salafis’ aversion to nationalism: “Muslim nationalists. Your beloved nationalistic countries weren’t even on the map a hundred years ago. There’s no magical guarantee that they’ll stick around forever. It’s time to embrace the power of Islamic ideology instead of obsessing over those silly imaginary lines on a globe”. The tweet was correctly identified as Salafi because its content embodies Salafi thought. Most Salafis view nationalism as inherently conflicting with the idea of tawhid, the oneness of God. For Salafis, believing in the oneness of God entails, first accepting God as the sole creator and sovereign of the universe (tawhid al-rububiyya), and, second, worshiping only Him (tawhid al-uluhiyya). An integral part of tawhid al-uluhiyya is the belief that Allah alone has the exclusive right to legislate (what Salafi-jihadis label as tawhid al-hakimiyya). Hence, for Salafis, if a Muslim recognizes the nation as a legitimate political entity, he must adhere to its laws. This will inevitably create a conflict of loyalty because state law and divine law are incompatible for two reasons. First, a nation may, and often does, legislate laws that contradict divine law. This was the case, for example, with the Tunisian law that banned polygamy in 1956, thus sidelining the divine permission to marry multiple wives.15 Second, nationalism assumes that citizens of a state have certain obligations of loyalty and/or fraternity toward one another, regardless of the citizens’ religious affiliation. An Egyptian Muslim must be loyal to and protect his fellow Egyptian Copt. This, however, conflicts with the Salafi doctrine of al-Wala’ wal-Bara (Loyalty and Disavowal), which confines Muslims’ loyalty to other Muslims, regardless of their citizenship, and prohibits them from being loyal even to their non-Muslim fellow citizens.

A tweet about anti-rationalism, flagged by the LSTM model as Salafi, reflects a profound aspect of Salafi ideology: “Shaykh al-Islam Ibn Taymiyyah on the Role of Reason in Guidance in Matters of Creed and Clarification of a Doubt”. Ibn Taymiyya and the Salafis maintain that rational thinking is not guaranteed to lead a person to the truth in matters of creed.16 They cite the Muslim theologians (mutakallimun) as an example of people who pursued rational thinking but were not able to grasp the simple truth that God is above the heavens because “they [were] blinded by Aristotelian Metaphysics”. Salafis maintain that matters of creed cannot always be understood or proven by logic, and the flagged tweet reflects this conviction (Abu Iyaad 2024).

Consider another tweet whose title includes what can be defined as second-tier Salafi keywords: “Madakhila are Jamiyya”. The Jamiyya, or the Jami movement, were a group of scholars in Saudi Arabia who opposed the Sahwa movement in the 1990s. The Sahwa leaders (e.g., Safar al-Hawali and Salman al-ʻAwda) questioned the Saudi regime’s legitimacy because it did not rule exclusively and completely in accordance with Shari‘a law. In response to the Sahwa criticism, members of the Jami movement, including among others, Shaykh Rabiʻ al-Madkhali, attacked the Sahwa for being involved in politics instead of focusing on rectifying society religiously. In turn, members of the Sahwa movement portrayed Jami scholars as members of the “party of the rulers” and chastised them for siding with the regime instead of fulfilling their obligation to ensure that society is ruled by divine law. The comparison of Madakhila to Jamiyya in the short tweet cited above thus incorporates a term and an idea that make sense only when one understands Salafi thinking (Alshech 2023, p. 303).

Finally, the tweet cited above, “Allah judges with truth while those whom they invoke besides him cannot judge anything”, reflects the Salafi refusal to rule and judge by any law other than the Qur’an and the Prophetic tradition. Salafis rebuke Muslim societies for accepting constitutions that combine both Shari‘a law and foreign law, claiming that only Allah can guide to the truth, while other human legislators, which are, by definition, fallible, can only lead people astray. The LSTM model identified the tweet above based on a connection between the idea expressed in the tweet and Salafi beliefs.

These examples demonstrate that the LSTM model is able to identify Salafi content online even when the text expressing it is devoid of patently Salafi keywords. At this time, most of the tweets collected by the system do contain Salafi keywords, but the fact that many do not indicates that the system is able to learn and identify Salafi ideology. Looking forward, it is anticipated that the more the model flags keyword-free Salafi content, the more the model can improve its ability to locate subtle and sometimes elusive Salafi content online.

7.4. Tweets That Reflect Salafi Trends Within the UK

In addition to correctly identifying as Salafi tweets that lack evident Salafi keywords, the model is also valuable because its identification enables researchers to track UK Salafis’ affiliations with the various existing Salafi trends. Consider, for example, the following tweet: “[N]ew video: Islam against extremism-shaykh [H]asan al-Somali”.17 In his lecture, Hasan al-Somali expresses negative attitudes towards the Salafi-jihadi trends manifested in groups such as ISIS, Boko Haram, al-Qaeda, and Shabab al-Mujahidin. Al-Somali also speaks against the murjiʼa. The murjiʼa was a seventh century movement that withheld judgment on whether the third and fourth caliphs (i.e., ʻUthman and ʻAli) were believers.18 Modern Salafis use the label murjiʼa pejoratively to describe contemporary Muslims, like the Muslim Brothers,19 who refrain from proclaiming takfir against co-religionists (Lav 2012, p. 41ff) who are negligent in their religious practice (Al-Masri 1993, p. 41; Al-Najdi 2012, p. 1:187).20 The tweet promoting al-Somali’s video appears to be compatible with the views of traditional Salafis (Purists) who reject both jihadi tendencies (e.g., Salafi-jihadism) and an anti-takfir proclivity (e.g., Muslim Brothers). Understanding which trends are prevalent in which areas in the United Kingdom can contribute significantly to contemporary research.

Another tweet similarly reflects the negative attitude of Salafis towards the Muslim Brothers: “The Aqidah of the Muslim Brotherhood- by Hakeem Bilal Davis”.21 The tweet references a video lecture by Davis in which he blames the Muslim Brothers for being what he describes as jama‘at al-tajmi‘iyya, Muslim movements that gather Muslims for the sole purpose of establishing an Islamic Caliphate. He claims that, in order to unite Muslims around the idea of a Caliphate, these groups are willing to adopt a superficial understanding of the first declaration of belief (“there is no god besides Allah”). For the Muslim Brothers, he claims, the phrase means that “there is no deity but for Allah”. But, according to Salafism, the statement of belief means that no one has the right to be worshiped but Allah, and no deity other than Allah may legislate. Like the previous tweet, this one too reflects opposition to the Muslim Brothers. Taken together, it is possible that such tweets may reflect the current existence of dogmatic Salafi tendencies in the United Kingdom. To the extent such tendencies are in fact widespread within the United Kingdom, they are reminiscent of the early Salafi phase in the United Kingdom represented by the Ahl al-Hadith and thus may evidence a withdrawal among UK Salafis from the idea promoted by the JIMAS organization (see discussion above), which toned down its hardline criticism of other Islamic denominations particularly after 9/11 and 7/7 attacks. Additional research is required to determine whether this is in fact the current trend in the UK.

The LSTM model flagged several tweets reflecting the debate between the Salafi Purists and Politicos in Saudi Arabia (as discussed by Wiktorowicz above). One such tweet is “The Madkhali myth. What is a so-called Madkhali? Shaykh Salih al-Fawzan and other senior scholars have stated it is merely a derogatory term with no tangible reality”. As stated above, the debate between the Purists and Politicos was tied to the Sahwa movement’s questioning of the Saudi regime’s legitimacy based on the regime’s failure to implement the Shari‘a fully and exclusively. The British author of this tweet seems to side with OASIS, the pro-Purist British Salafi movement, which criticized JIMAS, another splinter Salafi group in the UK, for the JIMAS’s support of the Politicos. He explains, further down in the tweet, that the title Madkhali, which initially was bestowed on followers of shaykh Rabi‘ al-Madkhali in Saudi Arabia, became a derogatory term attached by pro-Sahwa Salafis (Politicos) to any scholar who advocates obedience to contemporary rulers, even in matters that are not permitted by Islamic law, and to people who receive salaries from the ruler. The tweet’s author appears to side with the Purists who view protests and public criticism of the regime as prohibited.

Another tweet that reflects criticism of the Politicos is the following: “The Evil Effects of Bidah”. The tweet cites a lecture by Abu Khadeejah, a local Salafi scholar in the UK, in which he condemns the idea of promoting a political change by way of demonstrations, revolution, or protest, as the Politicos did in Saudi Arabia. In his view, these are political devices that were not known or practiced during the time of the Prophet Muhammad and accordingly are forbidden innovations.22

In short, primary results indicate that the system’s identification of tweets as Salafi is useful to researchers not only in identifying the Salafi virtual community in the UK but also in discerning the various ideological trends that exist within that community. The more data the system accumulate, the more that researchers will be able to identify and trace sub-communities based on their adoption of specific Salafi positions regarding questions that divide the global Salafi world today.

8. Conclusions

Machine learning (which is part of AI) is an essential tool in attempting to study any virtual community. In order to apply such methods to the virtual Salafi community in the UK, it became necessary to develop a model capable of identifying popular Salafi communications on social media. To this end, experts in computer science, Middle Eastern Studies, and Salafi Islam joined forces and engaged in expert–expert and machine–expert dialogue, which ultimately led to the creation of a system that flags Salafi writings with high accuracy. The models described here have several important advantages. First, they enable researchers to identify Salafi tweets that would not otherwise be identified based only on keywords. This increases the amount of raw data available to researchers working online. Second, they enable researchers to geographically map the various virtual Salafi communities across the UK. Finally, the models enable the identification of virtual Salafi sub-communities classified by their views on some of the prominent questions that divide Salafis worldwide. While the machine learning process described here does not replace experts in Islam, it does help such experts conduct effective research in cyberspace where data are abundant, but content is often elusive. The automated model described here discovers and accumulates the data that experts in Islam and Salafi-Islam then analyze manually.