Detecting Online Hate Speech Using Context Aware Models

Abstract

In the wake of a polarizing election, the cyber world is laden with hate speech. Context accompanying a hate speech text is useful for identifying hate speech, which however has been largely overlooked in existing datasets and hate speech detection models. In this paper, we provide an annotated corpus of hate speech with context information well kept. Then we propose two types of hate speech detection models that incorporate context information, a logistic regression model with context features and a neural network model with learning components for context. Our evaluation shows that both models outperform a strong baseline by around 3% to 4% in F1 score and combining these two models further improve the performance by another 7% in F1 score.

Challenges for Toxic Comment Classification: An In-Depth Error Analysis

“…The task of toxic comment classification lacks a consistently labeled standard dataset for comparative evaluation (Schmidt and Wiegand, 2017). While there are a number of annotated public datasets in adjacent fields, such as hate speech (Ross et al, 2016;Gao and Huang, 2017), racism/sexism (Waseem, 2016;Waseem and Hovy, 2016) or harassment (Golbeck et al, 2017) detection, most of them follow different definitions for labeling and therefore often constitute different problems.…”

“…Bidirectional GRU with Attention Layer. Gao and Huang (2017) phrase that "attention mechanisms are suitable for identifying specific small regions indicating hatefulness in long comments". In order to detect these small regions in our comments, we add an attention layer to our bidirectional GRU-based network following the work of Yang et al (2016).…”

Cross-Domain Detection of Abusive Language Online

“…A bewildering plethora of different types of abusive language can be found online. Some of the types dealt with in related work include but are not limited to sexism, racism (Waseem and Hovy, 2016;Waseem, 2016), toxicity (Kolhatkar et al, 2018), hatefulness (Gao and Huang, 2017), aggression (Kumar et al, 2018), attack (Wulczyn et al, 2017), obscenity, threats, and insults. A typology of abusive language detection subtasks was recently proposed by Waseem et al (2017).…”

“…Traditional machine learning approaches to detecting abusive language include the naive Bayes classifier (Kwok and Wang, 2013;Chen et al, 2012;Dinakar et al, 2011), logistic regression (Waseem and Hovy, 2016;Wulczyn et al, 2017;Burnap and Williams, 2015), and support vector machines (SVM) Dadvar et al, 2013;Schofield and Davidson, 2017). The best performance is most often attained by deep learning models, the most popular being convolutional neural networks (Gambäck and Sikdar, 2017;Potapova and Gordeev, 2016;Pavlopoulos et al, 2017) and variants of recurrent neural networks (Pavlopoulos et al, 2017;Gao and Huang, 2017;Pitsilis et al, 2018;Zhang et al, 2018). Some approaches (Badjatiya et al, 2017;Park and Fung, 2017;Mehdad and Tetreault, 2016) also rely on combining different types of models.…”

Preemptive Toxic Language Detection in Wikipedia Comments Using Thread-Level Context

“…Many varieties of toxic language have been considered in NLP research, including sexism, racism (Waseem and Hovy, 2016a;Waseem, 2016), toxicity (Kolhatkar et al, 2018), hatefulness (Gao and Huang, 2017a), aggression (Kumar et al, 2018), attack (Wulczyn et al, 2017a), obscenity, threats, and insults. Waseem et al (2017) proposed a systematic typology of toxic language.…”