A Robust Ensemble Learning Model for Fine-Grained Detection of Cyber Harassment

Abstract

Due to innovative growth in cyberspace and portable penetration, kids in India are at risk of cyber harassment. A review of 174 central graders in Delhi exposed that a total of 8% pandered in cyber harassment and 17% described being offended by such acts. However, the occurrence of in-person harassment, hostility, and discrimination is also happening. Amid all the digital platforms, Instagram positions developed with extreme cyber harassment. Over the preceding span, it brought substantial developments in the grounds of machine learning (ML) which have been efficaciously functional in fields associated with cyber harassment findings, such as buzz recognition, sentimentality study, and forged broadcast discovery. In machine learning, methods are provided for detective work and the classification of online harassment. Researchers have conducted experiments using comparable datasets. To build an ensemble learning model for detecting and classification different categories of online harassment from social media platforms. In our proposed work, a robust method of detecting online harassment (cyberbullying) on the Instagram dataset is used. The attributes of abusive words are initially analyzed from feature selection and pre-trained word embedding language models like BERT and ELMO. The harassment words are detected using unsupervised machine learning techniques such as association rule classifier, latent semantic analysis (LSA), and clustering technique. Then, a novel ensemble typical model is planned for categorizing the different types of online harassment using extreme gradient boosting (XGBoost) learning method. Hence, our robust method of ensemble model for detecting and classification of online harassment provides much better results with high accuracy and lesser loss function.