Cyber Hate Speech on Twitter: An Application of Machine Classification and Statistical Modeling for Policy and Decision Making

By P. Burnap, Matthew Leighton Williams. 2015

631 citations recorded by the source corpus.

The use of “Big Data” in policy and decision making is a current topic of debate. The 2013 murder of Drummer Lee Rigby in Woolwich, London, UK led to an extensive public reaction on social media, providing the opportunity to study the spread of online hate speech (cyber hate) on Twitter. Human annotated Twitter data was collected in the immediate aftermath of Rigby's murder to train and test a supervised machine learning text classifier that distinguishes between hateful and/or antagonistic responses with a focus on race, ethnicity, or religion; and more general responses. Classification features were derived from the content of each tweet, including grammatical dependencies between words to recognize “othering” phrases, incitement to respond with antagonistic action, and claims of well-founded or justified discrimination against social groups. The results of the classifier were optimal using a combination of probabilistic, rule-based, and spatial-based classifiers with a voted ensemble meta-classifier. We demonstrate how the results of the classifier can be robustly utilized in a statistical model used to forecast the likely spread of cyber hate in a sample of Twitter data. The applications to policy and decision making are discussed.

Subfields: Artificial Intelligence, Sociology and Political Science, Natural Language Processing, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Sociology and Political Science

In Natural Language Processing

In Safety, Robustness & Interpretability

Citations in the corpus

HateCheck: Functional Tests for Hate Speech Detection Models (2020, background): “This is motivated by previous work, which highlights syntactic relationships and long-range dependencies as model weak points (Burnap and Williams, 2015; Vidgen et al., 2019).”
The Risk of Racial Bias in Hate Speech Detection (2019, background): “Many datasets have been created, but most are either small-scale pilots (∼100 instances; Kwok and Wang, 2013; Burnap and Williams, 2015; Zhang et al., 2018), or focus on other domains (e.”
Language Generation Models Can Cause Harm: So What Can We Do About It? An Actionable Survey (2022, neutral): “Application Level Interventions Feature-based Detection Toxicity Lexical features (Xiang et al., 2012; Dadvar et al., 2012; Burnap and Williams, 2015; Liu and Forss, 2015); n-gram features (Chen et al.”
Automated Hate Speech Detection and the Problem of Offensive Language (2017, methodology): “Other supervised approaches to hate speech classification have unfortunately conflated hate speech with offensive language, making it difficult to ascertain the extent to which they are really identifying hate speech (Burnap and Williams 2015; Waseem and Hovy 2016).”
Racial Bias in Hate Speech and Abusive Language Detection Datasets (2019, background): “A number of studies have shown that false positive cases of hate speech are associated with the presence of terms related to race, gender, and sexuality (Kwok and Wang, 2013; Burnap and Williams , 2015; Davidson et al., 2017).”
A Survey on Hate Speech Detection using Natural Language Processing (2017, methodology): “Burnap and Williams (2015) apply some statistical feature selection ( Bayesian Logistic Regression ), Chen et al. (2012) and Gi-tari et al. (2015) manually select the relations (e.g. by enforcing that one argument of the relation is an offensive term) while Nobata et al. (2016) do not carry out any…”
Hate Speech Dataset from a White Supremacy Forum (2018, methodology): “The approach presented in this paper follows (Nobata et al., 2016; Burnap and Williams, 2015; Djuric et al., 2015) (i.”
Challenges for Toxic Comment Classification: An In-Depth Error Analysis (2018, methodology): “Toxic comment identiﬁcation is a supervised clas-siﬁcation task and approached by either meth-ods including manual feature engineering (Burnap and Williams, 2015; Mehdad and Tetreault, 2016; Waseem, 2016; Davidson et al., 2017; Nobata et al., 2016; Kennedy et al., 2017; Samghabadi et al., 2017;…”
Understanding Abuse: A Typology of Abusive Language Detection Subtasks (2017, background): “Previous work has identiﬁed instances of hate speech that are both directed and generalized (Burnap and Williams, 2015; Waseem and Hovy, 2016; Davidson et al., 2017), although Nobata et al. (2016) come closest to making a distinction between directed and generalized hate.”
SemEval-2019 Task 6: Identifying and Categorizing Offensive Language in Social Media (OffensEval) (2019, background): “Hate speech identification: This is the most studied abusive language detection task (Kwok and Wang, 2013; Burnap and Williams, 2015; Djuric et al., 2015).”
Analyzing the Targets of Hate in Online Social Media (2016, background): “Existing eﬀorts are focused on studying posts of known hate groups or radical forums [1, 2, 5, 7, 10].”
Predicting the Type and Target of Offensive Posts in Social Media (2019, methodology): “, 2017), only 5% of the tweets were considered hate speech by the majority of the annotators, and in (Burnap and Williams, 2015) only 11.”
All You Need is "Love": Evading Hate Speech Detection (2018, background): “Of linguistic features applied to hate speech classification, the most common have been part-of-speech tags and dependency relations [5, 6, 20].”
Fighting Offensive Language on Social Media with Unsupervised Text Style Transfer (2018, methodology): “Most previous work that address the problem of offensive language on social media has focused on text classification using different machine learning methods (Xiang et al., 2012; Warner and Hirschberg, 2012; Kwok and Wang, 2013; Wang et al., 2014; Burnap and Williams, 2015; Nobata et al., 2016; Davi”
Generating Counter Narratives against Online Hate Speech: Data and Strategies (2020, background): “…2012; Silva et al., 2016; Schmidt and Wiegand, 2017; Fortuna and Nunes, 2018) employing features such as lexical resources (Gitari et al., 2015; Burnap and Williams, 2016), sentiment polarity (Burnap and Williams, 2015) and multi-modal information (Hosseinmardi et al., 2015) to a classiﬁer.”
CONAN - COunter NArratives through Nichesourcing: a Multilingual Dataset of Responses to Fight Online Hate Speech (2019, background): “The predominant approaches are to build a classiﬁer trained on various features derived from lexical resources (Gitari et al., 2015; Burnap and Williams, 2015, 2016), n-grams (Sood et al., 2012; Nobata et al., 2016) and knowledge base (Dinakar et al., 2012), or to utilize deep neural networks…”
Implicitly Abusive Language – What does it actually look like and why are we not getting there? (2021, background): “Previous work already established that the deﬁnite article, which represents a subset of such generic phrases, is predictive for abusive language (Burnap and Williams, 2015; Palmer et al., 2017).”