Introducing the Gab Hate Corpus: Defining and applying hate-based rhetoric to social media posts at scale

Kennedy, Brendan; Atari, Mohammad; Am, Davani; L, Yeh; Omrani, Ali S.; Kim, Yonggyun; Coombs, Kris; Havaldar, Shreya; Portillo-Wightman, Gwenyth; González, E. Herrera; Hoover, Jacob Louis; Azatian, A; Hussain, Amir; Lara, Antonio Jiménez; g, olmos; Omary, A; Park, C; Wang, C; Wang, X; Zhang, Yujiao; Dehghani, Morteza

doi:10.31234/osf.io/hqjxn

Cited by 38 publications

(44 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Lastly, especially in the purely predictive setting, these feature-based methods are best used as baselines for more sophisticated models. For example, Kennedy, Atari, Mostafazadeh Davani, Yeh, et al, (2020) compared a feature-based model (using TF-IDF vectors) to a leading method in NLP for predicting whether social media posts contained hate-based rhetoric. Table 4 .…”

Section: Methods For Feature-based Supervisionmentioning

confidence: 99%

Text Analysis for Psychology: Methods, Principles, and Practices

Kennedy¹,

Ashokkumar²,

Boyd³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Due to the explosion of new sources of human language data and the rapid progression of computational methods for extracting meaning from natural language, language analysis is a promising, though complicated, category of psychological research. In this chapter, we give a modern perspective on language analysis as it applies to psychology, uniting historical context, the diverse range of domains studied in psychology via language, and the methodological rigor of natural language processing (NLP) and machine learning. Top–down methods (e.g., dictionary approaches, text annotation) are presented alongside bottom–up methods (e.g., topic modeling, word embedding, language modeling) in order to give the reader a comprehensive grounding in the tools available and the recommended practices involved in applying them. We conclude with a view of the future of language analysis, specifically the ways in which psychology and NLP will continue to co-develop.

show abstract

Section: Methods For Feature-based Supervisionmentioning

confidence: 99%

Text Analysis for Psychology: Methods, Principles, and Practices

Kennedy¹,

Ashokkumar²,

Boyd³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Given this, a precision score of 0.75 (one of four messages flagged as hate speech will be done so incorrectly), represents a reasonable level of performance, and favourable in relation to comparable models (e.g. Davidson et al, 2017;Kennedy et al, 2020).…”

Section: Automatic Detection Of Hate Speech In Social Media Postsmentioning

confidence: 97%

Hate Contagion: Measuring the spread and trajectory of hate on social media

Jd¹,

Bright²

2021

Preprint

View full text Add to dashboard Cite

Online hate speech is a growing concern, with minorities and vulnerable groups increasingly targeted with extreme denigration and hostility. The drivers of hate speech expression on social media are unclear, however. This study explores how hate speech develops on a fringe social media platform popular with the far-right, Gab. We investigate whether users seek out this platform in order to express hate, or whether instead they develop these opinions over time through a mechanism of socialisation, as they interact with other users on the platform. We find a positive association between the time users spend on the platform and their hate speech expression. We show that while some users do arrive on these platforms with pre-existing hate stances, others develop these expressions as they get exposed to the hateful opinions of others. Our analysis reveals how hate speech develops online, the important role of the group environment in accelerating its development, and gives valuable insight to inform the development of counter measures.

show abstract

“…Prediction Task Bias GHC (Kennedy et al, 2018) Hate Group Identifier Stormfront (de Gibert et al, 2018) Hate Group Identifier DWMW (Davidson et al, 2017) Toxicity AAVE Dialect FDCL (Founta et al, 2018) Toxicity AAVE Dialect BiasBios (De-Arteaga et al, 2019) Occupation Gender Stereotyping OntoNotes 5.0 (Weischedel et al, 2013) Coreference Gender Stereotyping fore. This is a key test of UBM's viability for widespread application.…”

Section: Datasetmentioning

confidence: 99%

“…This bias refers to higher false positive rates of hate speech predictions for sentences containing specific group identifiers, which is harmful to protected groups by misclassifying innocuous text (e.g., "I am a Muslim") as hate speech. We include two datasets for study, namely the Gab Hate Corpus (GHC; Kennedy et al, 2018) and the Stormfront corpus (de Gibert et al, 2018). Both datasets contain binary labels for hate and non-hate instances, though with differences in the labeling schemas and domains.…”

Section: Bias Factors and Datasetsmentioning

confidence: 99%

“…For explanation regularization algorithm, we set the regularization strength α as 0.03 for GHC and Stormfront experiments, and 0.1 for FDCL and DWMW experiments. We regularize importance score on 25 group identifiers in (Kennedy et al, 2018) for GHC and Stormfront. These group identifiers the ones that have the largest coefficient in a bag-of-words linear classifier.…”

Section: A Implementation Detailsmentioning

confidence: 99%

See 1 more Smart Citation

On Transferability of Bias Mitigation Effects in Language Model Fine-Tuning

Jin¹,

Barbieri²,

Kennedy³

et al. 2021

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Langua

Self Cite

View full text Add to dashboard Cite

Fine-tuned language models have been shown to exhibit biases against protected groups in a host of modeling tasks such as text classification and coreference resolution. Previous works focus on detecting these biases, reducing bias in data representations, and using auxiliary training objectives to mitigate bias during fine-tuning. Although these techniques achieve bias reduction for the task and domain at hand, the effects of bias mitigation may not directly transfer to new tasks, requiring additional data collection and customized annotation of sensitive attributes, and re-evaluation of appropriate fairness metrics. We explore the feasibility and benefits of upstream bias mitigation (UBM) for reducing bias on downstream tasks, by first applying bias mitigation to an upstream model through fine-tuning and subsequently using it for downstream finetuning. We find, in extensive experiments across hate speech detection, toxicity detection, occupation prediction, and coreference resolution tasks over various bias factors, that the effects of UBM are indeed transferable to new downstream tasks or domains via finetuning, creating less biased downstream models than directly fine-tuning on the downstream task or transferring from a vanilla upstream model. Though challenges remain, we show that UBM promises more efficient and accessible bias mitigation in LM fine-tuning. 12

show abstract

Introducing the Gab Hate Corpus: Defining and applying hate-based rhetoric to social media posts at scale

Cited by 38 publications

References 34 publications

Text Analysis for Psychology: Methods, Principles, and Practices

Text Analysis for Psychology: Methods, Principles, and Practices

Hate Contagion: Measuring the spread and trajectory of hate on social media

On Transferability of Bias Mitigation Effects in Language Model Fine-Tuning

Contact Info

Product

Resources

About