Toxic Comment Detection Using Machine Learning

Abstract: The proliferation of online platforms and social media has led to a surge in abusive and disrespectful content, commonly known as toxic comments, which negatively impacts user engagement and mental well-being. Traditional moderation methods struggle with the sheer volume and linguistic complexity of this content. To address this challenge, Deep Learning (DL) models have emerged as highly effective tools for building intelligent and adaptive toxic comment detection systems. This research focuses on applying deep neural networks, particularly Bi-directional Long Short-Term Memory (BiLSTM) and Transformer models (e.g., BERT), to enhance the accuracy and multi-label classification of toxic language. The study involves extensive text preprocessing, advanced word embedding techniques (like Word2Vec and FastText), and training the DL models on large-scale datasets such as the Jigsaw Toxic Comment Classification Challenge. Experimental results demonstrate that the proposed BiLSTM- and BERT-based models achieve superior performance in classifying multiple types of toxicity (e.g., toxic, severe-toxic, threat, insult) compared to conventional machine learning approaches. This work highlights the potential of sophisticated Deep Learning architectures in providing robust and scalable solutions for fostering a safer online environment. Keywords: Toxic Comment Detection, Mechine learning.