Cyberbullying severity detection: A machine learning approach

Area of Science:

• Computer Science
• Social Computing
• Natural Language Processing

Background:

• Online social networks offer benefits but also expose users to cyberbullying and harassment.
• Detecting and categorizing the severity of cyberbullying is crucial for user safety.

Purpose of the Study:

• To propose a novel cyberbullying detection framework for Twitter content.
• To develop a supervised machine learning model for classifying cyberbullying and its severity.

Main Methods:

• Leveraged pointwise mutual information (PMI) to generate features from Twitter content.
• Utilized Embedding, Sentiment, Lexicon features, and PMI-semantic orientation.
• Applied machine learning algorithms including Naïve Bayes, KNN, Decision Tree, Random Forest, and Support Vector Machine.

Main Results:

• The proposed framework demonstrated promising results in both multi-class and binary cyberbullying detection settings.
• Achieved high performance metrics including Kappa, classifier accuracy, and f-measure.
• Experimental findings confirmed the significance of the proposed features for effective cyberbullying detection.

Conclusions:

• The developed framework provides a feasible and effective solution for detecting cyberbullying and its severity on social networks.
• The proposed features significantly enhance the performance of machine learning models in cyberbullying detection.
• This research contributes to creating safer online environments through improved cyberbullying identification.