Attention-based Hybrid Deep Learning Model for Detecting Hateful Tweets
Anchal Rawat, Santosh Kumar, Surender Singh Samant · 2024
The spread of hate speech on the internet makes it difficult to create safe online groups. Because harmful language is ever-evolving and has subtleties and contextual dependencies, automated hate speech identification is an intricate task. Identifying hate speech is essential to preserving a secure online environment. We introduce a hybrid neural network architecture that integrates Bidirectional Long Short-Term Memory (BiLSTM) networks, Convolutional Neural Networks (CNNs), and attention mechanisms to enhance hate speech identification. While CNNs are good at capturing local n-gram characteristics, BiLSTMs are better at modeling long-range dependencies in text sequences. The attention mechanism dynamically highlights the most prominent textual portions that appear to be hate speech. Our experimental evaluation of the Davidson dataset shows that the proposed hybrid model achieves greater accuracy in identifying offensive language, hate speech, and neutral content compared to baseline techniques. Our model reaches a noteworthy 93% accuracy. This study advances the creation of more reliable and comprehensible instruments for reducing hate speech online.