Machine Learning Enhanced By Sentiment Analysis For Cyberbullying Detection Using Nlp And Lstm

Abstract

This research presents an automated deep learning-based methodology for detecting cyberbullying in social media text, emphasizing the practical application of the LSTM architecture combined with Natural Language Processing (NLP). Safe digital communication relies heavily on the quality and efficiency of content moderation, making accurate cyberbullying detection critical for minimizing emotional harm and improving online experiences. Traditional manual moderation methods are time-consuming, inconsistent, and prone to human error, limiting scalability in large platforms. By leveraging memory cells and gating mechanisms of LSTM, the proposed system effectively extracts complex sequential features from textual data. This enables precise differentiation between offensive and non-offensive text while maintaining computational efficiency, allowing the model to function effectively even in environments with dynamic slang and linguistic nuances. A balanced and well-curated dataset of social media text was used to train and validate the LSTM-based model. The dataset includes diverse abusive types, capturing variations in harassment, insults, and threats. The model undergoes rigorous preprocessing, including tokenization, stop word removal, and lemmatization, to improve generalization and reduce overfitting. Experimental results demonstrate that the LSTM model achieves high classification accuracy while maintaining a lightweight footprint suitable for deployment in digital scenarios. Performance metrics confirm the model's reliability in identifying offensive text, highlighting its potential to enhance automated quality control in social media moderation.