Offensive Text Detection using Deep Learning: A Review with Open Challenges

Abstract

The rapid expansion of online communication platforms has led to an increase in offensive and harmful content, necessitating robust detection mechanisms. Deep learning techniques have emerged as a powerful approach for offensive text detection, leveraging neural networks to capture complex linguistic patterns and contextual nuances. This paper provides a comprehensive review of deep learning-based methods for offensive text detection, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory (LSTM), gated recurrent units (GRU), and transformer-based models such as BERT and GPT. We discuss key challenges such as class imbalance, context understanding, bias mitigation, and the interpretability of deep learning models. Furthermore, we explore evaluation metrics, benchmark datasets, and recent advancements in adversarial robustness and explainability. Finally, we highlight future research directions to improve the effectiveness, fairness, and scalability of deep learning techniques in offensive text detection