Hate speech detection: Challenges and solutions

By Sean MacAvaney, Hao-Ren Yao, Eugene Yang, Katina Russell, Nazli Goharian, Ophir Frieder. 2019

568 citations recorded by the source corpus.

As online content continues to grow, so does the spread of hate speech. We identify and examine challenges faced by online automatic approaches for hate speech detection in text. Among these difficulties are subtleties in language, differing definitions on what constitutes hate speech, and limitations of data availability for training and testing of these systems. Furthermore, many recent approaches suffer from an interpretability problem—that is, it can be difficult to understand why the systems make the decisions that they do. We propose a multi-view SVM approach that achieves near state-of-the-art performance, while being simpler and producing more easily interpretable decisions than neural methods. We also discuss both technical and practical challenges that remain for this task.

Subfields: Artificial Intelligence, 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 Natural Language Processing

In Safety, Robustness & Interpretability

Citations in the corpus

ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detection (2022, background): “One key challenge is that, in contrast to explicit toxicity, implicit toxicity is not marked by the use of profanity or swearwords, is sometimes positive in sentiment, and is generally harder to detect or collect at scale (MacAvaney et al., 2019; Breitfeller et al., 2019).”
Multilingual Sentiment Analysis and Toxicity Detection for Text Messages in Russian (2021, background): “Up to this point, the identiﬁcation of hateful and offensive statements was performed within the framework of solving the problem of detecting and classifying hate speech [7], work on which continues to this day [4], [8], [9].”
The Psychological Well-Being of Content Moderators: The Emotional Labor of Commercial Moderation and Avenues for Improving Support (2021, background): “or lethal violence, such as murder, suicide, violent extremism [42], animal abuse, hate speech [92, 136], sexual abuse, child or revenge pornography [133], and more [26, 87, 131].”
Detecting weak and strong Islamophobic hate speech on social media (2018, background): “, who also find that SVM outperforms deep learning for hate speech detection (MacAvaney et al., 2019).”
ToxCCIn: Toxic Content Classification with Interpretability (2021, background): “In some cases, toxicity can be expressed in subtle or implicit ways, such as through sarcasm or metaphor (MacAvaney et al., 2019; Waseem et al., 2017).”
Contextualizing Hate Speech Classifiers with Post-hoc Explanation (2020, methodology): “Speciﬁcally, we use post-hoc explanation algorithms to interpret and modulate ﬁne-tuned language models like BERT (Devlin et al., 2018), which achieve state of the art performance on many hate speech detection tasks (MacAvaney et al., 2019; Mandl et al., 2019).”
Large Language Model Alignment: A Survey (2023, background): “may sometimes exhibit positive sentiment and is typically harder to detect or collect on a large scale (MacAvaney et al., 2019; Breitfeller et al., 2019).”
Fast, Accurate, and Healthier: Interactive Blurring Helps Moderators Reduce Exposure to Harmful Content (2020, background): “Many machine learning solutions are being proposed to automatically moderate content to the extent possible (Deniz et al. 2014; Wang et al. 2012; Ries and Lienhart 2014; MacAvaney et al. 2019; Schmidt and Wiegand 2017; Jur-gens, Chandrasekharan, and Hemphill 2019).”
Towards Agile Text Classifiers for Everyone (2023, background): “The underlying models for detecting these attributes have also grown in sophistication, from linear models such as support vector machines (MacAvaney et al., 2019) to more modern approaches that rely on deep learning architectures like CNNs (Gambäck and Sikdar, 2017) and, more recently,…”
HABERTOR: An Efficient and Effective Deep Hatespeech Detector (2020, methodology): “…hatespeech, researchers developed human-crafted feature-based classifiers (Chatzakou et al., 2017; Davidson et al., 2017; Waseem and Hovy, 2016; MacAvaney et al., 2019), and proposed deep neural network architectures (Zampieri et al., 2019; Gambäck and Sikdar, 2017; Park and Fung, 2017;…”
COLD: A Benchmark for Chinese Offensive Language Detection (2022, background): “Tremendous works are endeavoring to detect hate speech based on automatic identification like topic analysis and keyword-based detection [Warner and Hirschberg, 2012; MacAvaney et al., 2019].”