A Large Labeled Corpus for Online Harassment Research

By Jennifer Golbeck, Zahra Ashktorab, Rashad O. Banjo, Alexandra Berlinger, Siddharth Bhagwan, C. Buntain, Paul Cheakalos, Alicia A. Geller, Quint Gergory, R. Gnanasekaran and 20 more. 2017

232 citations recorded by the source corpus.

A fundamental part of conducting cross-disciplinary web science research is having useful, high-quality datasets that provide value to studies across disciplines. In this paper, we introduce a large, hand-coded corpus of online harassment data. A team of researchers collaboratively developed a codebook using grounded theory and labeled 35,000 tweets. Our resulting dataset has roughly 15% positive harassment examples and 85% negative examples. This data is useful for training machine learning models, identifying textual and linguistic features of online harassment, and for studying the nature of harassing comments and the culture of trolling.

Subfields: Artificial Intelligence, Information Systems, Social Psychology, 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 Information Systems

In Social Psychology

In Natural Language Processing

In Safety, Robustness & Interpretability

Citations in the corpus

HateCheck: Functional Tests for Hate Speech Detection Models (2020, background): “F5 – direct threats against a protected group or its members : Core element of several hate speech taxonomies (Golbeck et al., 2017; Zampieri et al., 2019; Vidgen et al., 2020a; Banko et al., 2020) F6 – threats expressed as normative statements : Highlighted by an interviewee as a way of avoiding…”
The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memes (2020, background): “Several text-only hate speech datasets have been released, mostly based on Twitter [83, 82, 11, 24, 20], and various architectures have been proposed for classifiers [45, 54, 53].”
Learning from the Worst: Dynamically Generated Datasets to Improve Online Hate Detection (2021, methodology): “This category is used in datasets by Hammer (2014), Golbeck et al. (2017) and Anzovino et al. (2018).”
Racial Bias in Hate Speech and Abusive Language Detection Datasets (2019, methodology): “The datasets considered here relied upon a range of different annotators, from the authors (Golbeck et al., 2017; Waseem and Hovy, 2016) and crowdwork-ers (Davidson et al., 2017; Founta et al., 2018) to activists (Waseem, 2016).”
Challenges and frontiers in abusive content detection (2019, background): “Creating appropriate datasets for training hate detection systems is a crucial but time-consuming task (Golbeck et al., 2017).”
Resources and benchmark corpora for hate speech detection: a systematic review (2020, background): “Golbeck et al. (2017) (GAB)—35,000 tweets in English annotated as ‘‘ harassing/ not harassing ’’ by 2 judges, plus a third one to settle cases in disagreement.”
Towards Ethics by Design in Online Abusive Content Detection (2020, background): “Golbeck et al. (2017) collected judgements from two trained annotators, and a third annotator was employed only if the ﬁrst two disagreed.”
Towards a Comprehensive Taxonomy and Large-Scale Annotated Corpus for Online Slur Usage (2020, background): “However, these methods conceal but do not remove systemic bias in the overall data (Gonen and Goldberg, 2019).”
Handling Bias in Toxic Speech Detection: A Survey (2022, background): “A case study is the large-scale harassment annotation code book [38] that helped label 35k instances of various categories of online harassment, including hate speech.”
Directions in abusive language training data, a systematic review: Garbage in, garbage out (2020, background): “[57] state that messages that are “unapologetically or intentionally offensive” fit in the highest grade of trolling under their schema.”
Directions in Abusive Language Training Data: Garbage In, Garbage Out (2020, background): “[50] find that messages that are “unapologetically or intentionally offensive" fit in the highest grade of trolling under their schema.”
Confronting Abusive Language Online: A Survey from the Ethical and Human Rights Perspective (2020, background): “Golbeck et al. (2017) collected judgements from two trained annotators, and a third annotator was employed only if the first two disagreed.”