NLP-Driven Detection of Cyber-Bullying Comments in Instagram Social Network

People's attention towards usage of social media is growing continuously, particularly youngsters are paying special attention by means of advancing their professions and expanding network. Cyber-bullying, includes posting derogatory comments on other people's posts, fabricating an identity, and disseminating embarrassing photos or videos, is also on the rise on social media. Victims of cyber-bullying may result in low self-esteem, increased suicidal thoughts, and other negative emotions like anxiety, frustration, fury, or sadness of the person. Undoubtedly numerous calculations like Machine Learning (ML) methods have been utilized in recognizing cyber-bullying exercises which remain totally frozen. The primary focus of the proposed work is to integrate Term Frequency-Inverse Document Frequency (TF-IDF) information retrieval method with a Random Forest (RF) classifier. TF-IDF decreases the dimensionality of the dataset by turning textual data into numerical features and concentrating on informative words. RF classifier handles the words with high TF-IDF scores in a corpus and classifies the comments as “Low” and “High” depending on the severity flagged as potential indicators of bullying behavior. This framework develops software which automatically identifies and prevents hateful comments and hides them from the victim's page. Compared to other classifiers, RF adapts many linguistic patterns, manages copious data generated on social platform and handles imbalances in cyberbullying comments, particularly when paired with class weighting or sampling approaches. This work mainly focusses on Instagram social media page where it accomplishes 88% of accuracy, 0.89 of precision, 0.94 of recall and 0.91 of F-measure in detection of harmful words.