Online Library

Online abusive content detection is an inherently difficult task. It has received considerable attention from academia, particularly within the computational linguistics community, and performance appears to have improved as the field has matured. However, considerable challenges and unaddressed frontiers remain, spanning technical, social and ethical dimensions. These issues constrain the performance, efficiency and generalizability of abusive content detection systems. In this article we delineate and clarify the main challenges and frontiers in the field, critically evaluate their implications and discuss potential solutions. We also highlight ways in which social scientific insights can advance research. We discuss the lack of support given to researchers working with abusive content and provide guidelines for ethical research.

This report aims to contribute to our understanding of online hate in the context of the requirements of the revised Audiovisual Media Services Directive (AVMSD) for Video Sharing Platforms (VSPs) to protect the general public from incitement to hatred or violence. However, online hate is complex and it can only be fully understood by considering issues beyond the very specific focus of these regulations. Hence, we draw on recent social and computational research to consider a range of points outside VSP regulations, such as the impact, nature and dynamics of online hate. For similar reasons,
we have considered expressions of hate across a range of online spaces, including VSPs as well as other online platforms. In particular, we have closely examined how online hate is currently addressed by industry, identifying key and emerging issues in content moderation practices. Our analyses will be relevant to a range of experts and stakeholders working to address online hate, including researchers, platforms, regulators and civil society organisations.

The aim of this thesis is to enhance our understanding of the nature and dynamics of Islamophobic hate speech amongst followers of UK political parties on Twitter. I study four parties from across the political spectrum: the BNP, UKIP, the Conservatives and Labour. I make three main contributions. First, I define Islamophobia in terms of negativity and generality, thus making a robust, theoretically-informed contribution to the study of a deeply contested concept. This argument informs the second contribution,
which is methodological: I create a multi-class supervised machine learning classifier for Islamophobic hate speech. This distinguishes between weak and strong varieties and can be applied robustly and at scale. My third contribution is theoretical. Drawing together my substantive findings, I argue that Islamophobic tweeting amongst followers of UK parties can be characterised as a wind system which contains Islamophobic hurricanes. This analogy captures the complex, heterogeneous dynamics underpinning Islamophobia on Twitter, and highlights its devastating effects. I also show that Islamist terrorist attacks drive Islamophobia, and that this affects followers of all four parties studied here. I use this finding to extend the theory of cumulative extremism beyond extremist groups to include individuals with mainstream affiliations. These contributions feed into ongoing academic, policymaking and activist discussions about Islamophobic hate speech in both social media and UK politics.

Islamophobic hate speech on social media is a growing concern in contemporary Western politics and society. It can inflict considerable harm on any victims who are targeted, create a sense of fear and exclusion amongst their communities, toxify public discourse and motivate other forms of extremist and hateful behavior. Accordingly, there is a pressing need for automated tools to detect and classify Islamophobic hate speech robustly and at scale, thereby enabling quantitative analyses of large textual datasets, such as those collected from social media. Previous research has mostly approached the automated detection of hate speech as a binary task. However, the varied nature of Islamophobia means that this is often inappropriate for both theoretically informed social science and effective monitoring of social media platforms. Drawing on in-depth conceptual work we build an automated software tool which distinguishes between non-Islamophobic, weak Islamophobic and strong Islamophobic content. Accuracy is 77.6% and balanced accuracy is 83%. Our tool enables future quantitative research into the drivers, spread, prevalence and effects of Islamophobic hate speech on social media.

The influence of social media on the spread of violent extremist narratives and online radicalisation processes has recently become a focal point for research in the fields of Preventing and Countering Violent Extremism; however, most of the studies thus far have focused on Western countries and have often been aimed at analysing phenomena such as homegrown and lone wolf terrorism or the online radicalisation of foreign terrorist fighters. Far less evidence-based research has explored the influence of social media on terrorism in Africa and even less regarding Mali in particular. In this report, ICCT and the United Nations Interregional Crime and Justice Research Institute (UNICRI) seek to fill this gap through an analysis of survey data from youth participants in their joint on-the-ground project in Mali—“Mali (Dis-) Engagement and Re-Integration related to Terrorism” (MERIT)—to determine the implications of social media use for violent extremism in Mali.

The online domain potentially provides research with a vast body of data. The big quest within contemporary research is to make sense of all this data. A possibility to handle such data is by combining methods from the fields of new media and linguistics. Several studies have sought to understand how radicalism online comes to exist and grows. To date, however, none of these studies have fruitfully analysed language patterns within online communities that move beyond keyword analysis. In this thesis, I demonstrate a proof of concept for a classifier analysing and predicting salient language features within online radical discourse from the social media platform Reddit. Data consist of two datasets, radical and non-radical in nature, both containing 1 millions lines of text per dataset. The radical dataset is known for its radical nature, promoting radicalism in a variety of beliefs such as anti-feminism or white supremacy. Using software libraries as NLTK and SciKit within Python, I submitted that data to keyword and collocation frequency count, lexical diversity, a part-of-speech tagger and ultimately as features for a document classifier. Results showed that the radical discourse used in this thesis contains salient language features and show a clear sense of a virtual community. Finally, I discuss the implications of this thesis and provide directions for further research. Data was provided by TNO The Hague as part of the VOX-Pol project.