Empowering Abusive Language Detection in Low-Resource Scenarios: A Dynamic LSTM-Based Abusive Content Detection with Attention Mechanism
Sanjib Bora, Nomi Baruah, Debajani Baruah, Swarnangka Barman, Mandira Neog · 2024
In India, with its rich linguistic landscape, abusive social media remarks that attack individuals or communities based on protected attributes provide serious difficulty. Research in this area is hampered by the lack of digital resources for low-resource Indian languages like Assamese, which exacerbates this complexity. The goal of this research is to close this gap by creating a system that can accurately identify abusive remarks made in Assamese on social media. The experiment produced promising results by leveraging the highlighted capabilities of deep learning techniques, namely Long Short-Term Memory (LSTM) networks enhanced with attention mechanisms. The accuracy of the LSTM model was 71.97%; however, by incorporating attention processes, performance was dramatically increased to 75.22%. The results demonstrate how well attention mechanisms and deep learning techniques work together to enhance the detection of abusive comments in Assamese on web-based platforms. Particularly given the limitations of low-resource languages, this research provides important insights meant to foster a more safe online environment.