A Generalizable Context-Aware Deep Learning Model for Abusive Language Detection

The proliferation of abusive language and hate speech in online content has become a pressing societal concern, necessitating effective detection methods. Recent years have witnessed a surge in datasets and computational methods for detecting abusive language, reflecting the growing interest in combating online abuse. Deep learning, in particular, has emerged as a powerful tool for addressing this pervasive issue. This paper presents a novel context-aware, attention-based Bidirectional Long Short-Term Memory (Bi-LSTM) model that relies exclusively on textual features. The model is designed for robust detection of abusive language. The proposed model integrates a domain-specific language model, HateBERT, with stacked Bi-LSTM and attention mechanism to enhance the processing capabilities of neural networks, enabling nuanced understanding of abusive language patterns. The versatility of the model is demonstrated through experiments on diverse abuse categories, showcasing its ability to effectively classify various types of abuse. The paper compares the model with existing state-of-the-art approaches and the findings underscore the potential of deep learning-based models in addressing the pervasive issue of online abusive behavior.