Deep Learning for Hate Speech Detection in Tweets

By Pinkesh Badjatiya, Shashank Gupta, Manish Gupta, Vasudeva Varma. 2017

1,283 citations recorded by the source corpus.

Hate speech detection on Twitter is critical for applications like controversial event extraction, building AI chatterbots, content recommendation, and sentiment analysis. We define this task as being able to classify a tweet as racist, sexist or neither. The complexity of the natural language constructs makes this task very challenging. We perform extensive experiments with multiple deep learning architectures to learn semantic word embeddings to handle this complexity. Our experiments on a benchmark dataset of 16K annotated tweets show that such deep learning methods outperform state-of-the-art char/word n-gram methods by ~18 F1 points.

Subfields: Artificial Intelligence

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

Citations in the corpus

Challenges for Toxic Comment Classification: An In-Depth Error Analysis (2018, background): “, 2018), related research includes the investigation of hate speech (Badjatiya et al., 2017; Burnap and Williams, 2016; Davidson et al., 2017; Gambäck and Sikdar, 2017; Njagi et al., 2015; Schmidt and Wiegand, 2017; Vigna et al., 2017; Warner and Hirschberg, 2012), online harassment (Yin and Davison”
Understanding Abuse: A Typology of Abusive Language Detection Subtasks (2017, background): “Word embeddings are also promising ways to capture terms associated with abuse (Djuric et al., 2015; Badjatiya et al., 2017), although they may still be insufficient for cases like 4Chan’s connotation of “skype” where a word has a dominant meaning and a more subversive one.”
What’s in the Box? An Analysis of Undesirable Content in the Common Crawl Corpus (2021, background): “…sources such as Twitter (Founta et al., 2018; Basile et al., 2019; Waseem and Hovy, 2016) and Reddit (Tadesse et al., 2019; Farrell et al., 2019), with both ML-based (Badjatiya et al., 2017) and count-based approaches (Davidson et al., 2017) achieving comparable results (Fortuna and Nunes, 2018).”
Detection of Abusive Language: the Problem of Biased Datasets (2019, methodology): “We need to ﬁnd an explanation for this deviation since Waseem is by far the most frequently used dataset for detecting abusive language (Badjatiya et al., 2017; Bourgonje et al., 2017; Pitsilis et al., 2018; Agrawal and Awekar, 2018; Karan and Snajder, 2018; Kshirsagar et”
One-step and Two-step Classification for Abusive Language Detection on Twitter (2017, methodology): “Badjatiya et al. (2017) experimented with deep learning-based models using ensemble gradient boost classifiers to perform multi-class classification on sexist and racist language.”
All You Need is "Love": Evading Hate Speech Detection (2018, methodology): “However, CNN+GRU trained on T2 was more resilient to all attacks than LSTM trained on T2, indicating that model choice also had an eﬀect. results from all attacks and their mitigation are presented in Table 8.”
Latent Hatred: A Benchmark for Understanding Implicit Hate Speech (2021, background): “Platforms have responded by banning hate groups and ﬂag-ging abusive text (Klepper, 2020), and the research community has developed increasingly competitive hate speech detection systems (Fortuna and Nunes, 2018; Badjatiya et al., 2017).”
Neural Character-based Composition Models for Abuse Detection (2018, methodology): “For the Twitter dataset, unlike previous research (Badjatiya et al., 2017; Park and Fung, 2017), we report the macro precision, recall, and F1 averaged over 10 folds of stratified CV (Table 1).”
Large Scale Crowdsourcing and Characterization of Twitter Abusive Behavior (2018, methodology): “A widely used and available dataset in this ﬁeld, is the one in (Waseem and Hovy 2016), employed by several studies working on hate speech detection such as in (Badjatiya et al. 2017), (Gamb¨ack and Sikdar 2017), (Park and Fung 2017), and (Jha and Mamidi 2017).”
Deep Learning and Medical Diagnosis: A Review of Literature (2018, neutral): “Deep learning has already found its application in transportation and greenhouse-gas emission control [68], traffic control [69], text classification [8,70], object detection [71], speech detection [72,73], translation [74] and in other fields.”
Aggression Identification Using Deep Learning and Data Augmentation (2018, background): “Previous work agrees that word n-grams are well-performing features for hate speech detection (Nobata et al., 2016; Badjatiya et al., 2017; Warner and Hirschberg, 2012; Davidson et al., 2017; Schmidt and Wiegand, 2017).”
BERT and fastText Embeddings for Automatic Detection of Toxic Speech (2020, positive): “Deep-learning techniques have shown to be very powerful in classifying toxic speech [7].”
Comparative Studies of Detecting Abusive Language on Twitter (2018, methodology): “Badjatiya et al. (2017) implemented Gradient Boosted Decision Trees classiﬁers using word representations trained by deep learning models.”
Toxicity Detection can be Sensitive to the Conversational Context (2021, methodology): “Such posts are called toxic (Borkan et al., 2019), offensive (Sap et al., 2020) or abusive (Thylstrup and Waseem, 2020), and systems detecting them (Waseem and Hovy, 2016; Pavlopoulos et al., 2017b; Badjatiya et al., 2017) are called toxicity (or offensive or abusive language) detection systems.”
Context Sensitivity Estimation in Toxicity Detection (2021, background): “These posts are called toxic (Borkan et al., 2019) or abusive (Thylstrup and Waseem, 2020), and systems detecting them (Waseem and Hovy, 2016; Pavlopoulos et al., 2017b; Badjatiya et al., 2017) are called toxicity (or abusive language) detection systems.”
Hierarchical CVAE for Fine-Grained Hate Speech Classification (2018, methodology): “Badjatiya et al. (2017) use the dataset provided by Waseem and Hovy (2016) to do threeway classification.”
Towards multidomain and multilingual abusive language detection: a survey (2021, background): “Recently, deep learning approaches have been applied, achieving state-of-the-art results for some languages [9, 78].”
Improving Hate Speech Detection with Deep Learning Ensembles (2018, methodology): “Most relevant to our experiments with neural networks and Twitter data are hybrid models (Badjatiya et al., 2017; Park and Fung, 2017) which combine outputs from different neural networks.”
Empirical Analysis of Multi-Task Learning for Reducing Model Bias in Toxic Comment Detection (2019, methodology): “For text classiﬁcation tasks, documents are usually vector-ized into bag-of-words (BoW) features (e.g. TF-IDF).”