Chapter 1 : Classification of Hate Speech Using Deep Neural Networks

In the Internet age where the information flow has grown rapidly, there is an increase in digital comm unication. The spread of hatred that was previously limited to ver bal communications has quickly moved over the Inter net. Social media and community forums that allow people to discuss a nd express their opinions are becoming platforms fo r the dissemination of hate messages. Many countries hav e de eloped laws to prevent online hate speech. The y hold the companies that run the social media responsible for thei failure to remove hate speech. However, manu al analysis of hate speech on online platforms is infeasible due to the huge amount of data as it is expensive and time co nsuming. Thus, it is important to automatically process the online user contents to detect and remove hate speech from onli ne media. Through this work, we propose some solutions for the proble m of automatic detection of hate messages. We perfo rm hate speech classification using embedding representations of w rds and Deep Neural Networks (DNN). We compare fas tText and BERT (Bidirectional Encoder Representations from Trans formers) embedding representations of words. Furthermore, we perform classification using two approaches: (a) us ing word embeddings as input to Support Vector Mach ines (SVM) and DNN-based classifiers; (b) fine-tuning of a BERT mod el for classification using a task-specific corpus. Among the DNNbased classifiers, we compare Convolutional Neural Networks (CNN), Bi-Directional Long Short Term Memory (BiLSTM) and Convolutional Recurrent Neural Network (CRNN) . The classification was performed on a Twitter da taset using three classes: hate, offensive and neither classes. Compared to the feature-based approaches, the BERT fine-tuning approach obtained a relative improvement of 16% in terms of macro-average F1-measure and 5.3% in terms of weighted F1-measure.