Automated Threat Detection in Social Media: A Review of Advanced Computational Techniques

Abstract

The widespread use of social media platforms has significantly amplified the volume and visibility of harmful digital content, including cyberbullying, hate speech, and threatening text. These threats not only endanger mental health and emotional well-being but also compromise public safety and digital harmony. This review paper provides an in-depth analysis of recent advancements in automated threat detection, focusing on the integration of hybrid Machine Learning (ML), Deep Learning (DL), and Natural Language Processing (NLP) techniques. The study examines a variety of models from traditional classifiers like SVM and Logistic Regression to modern transformer-based models such as BERT, RoBERTa, DistilBERT, and MuRIL. It highlights challenges in multilingual and low-resource language contexts, emphasizes the value of hyperparameter tuning and feature optimization, and explores methods for real-time deployment and model explainability. By synthesizing literature across multiple datasets, languages, and threat types, this review aims to guide future research and development of intelligent, ethical, and scalable systems for automated content moderation.

Keywords:  Threat Detection, Social Media, Hate Speech, Cyberbullying, NLP, Machine Learning, Deep Learning, Hybrid Models, Low-Resource Languages, BERT, Explainable AI