Detecting Cyberbullying in Social Media: An NLP-Based Classification Framework

Objectives: To accomplish better cyberbullying classification accuracy through Modified Term Frequency and Inverse Document Frequency (MTF-IDF) with Machine Learning (ML) technique. The cyberbullying classification is improved by modifying the parameter by hypertuning the concept in MTF-IDF using the Optuna method. Method: To categorize the bullying text, this study discusses the Ensemble model that integrates MTF-IDF and ML methods with sophisticated feature extraction approaches. This research has considered 47692 tweets along with the label of cyberbullying classes, and 1000 tweets is the sample size considered for testing proposed MTF-IDF with various ML algorithms. From the text data, MTF-IDF with ML has assisted in identifying the cyberbullying feature patterns for feature extraction by hyperparameter tuning the MTF-IDF method using Optuna technique for the accurate classification of cyberbullying tweets. Findings: In text classification, MTF-IDF with ML has assisted in identifying the cyberbullying feature patterns for feature extraction techniques by hyperparameter tuning the MTF-IDF method by Optuna technique for the accurate cyberbullying tweets classification. The proposed hyperparameter-tuned MTF-IDF is evaluated with traditional Natural Language Processing (NLP) methods by confusion matrix metrics like accuracy and balanced accuracy as 83.63% and 83.42%, respectively which are seen to be high when compared to traditional NLP methods. Novelty: The proposed framework focuses on cyberbullying detection in social media more precisely by text mining sentimental words using Modified NLP methods with hyperparameter tuning and improving the classification t using Optuna with various ML methods and modified entropy concept of TF-IDF as Modified TF-IDF (MTF-IDF).

Keywords: Social Media, Cyberbullying, Term Frequency ­ Inverse Document Frequency (TF­ IDF), Natural Language Processing, Machine Learning, Sentiment Analysis (SA), Optuna