Machine Learning Enhanced By Sentiment Analysis For Cyberbullying Detection Using NLP

Abstract

Cyberbullying has become a significant concern in the modern digital environment, negatively impacting individuals and society at large. Detecting harmful interactions on social media platforms is therefore essential, as these platforms represent a major channel of online communication. Conventional detection techniques relying on machine learning methods and pre-trained language models often struggle with computational overhead and limited capability to interpret subtle linguistic variations. To address these limitations, this study introduces an enhanced cyberbullying detection framework that integrates Natural Language Processing (NLP) techniques with Long Short-Term Memory (LSTM) networks.The proposed approach incorporates comprehensive text preprocessing procedures, including tokenization, stop-word elimination, stemming, and lemmatization, to generate clean and meaningful input data. Semantic and sentiment-based features are extracted using embedding strategies that preserve contextual relationships among words. These representations are subsequently processed by an LSTM model, which is capable of learning sequential dependencies and temporal patterns in textual data, thereby improving the detection of complex cyberbullying expressions.Furthermore, to mitigate class imbalance in multi-class classification scenarios, appropriate resampling strategies are applied, enhancing model stability and reducing bias. Experimental observations indicate that the integration of deep learning with structured NLP preprocessing improves both accuracy and contextual understanding, making the proposed framework effective for identifying cyberbullying content in online communications.