A Systematic Review of Social Media-Based Emotion and Threat Detection Techniques

Abstract

The proliferation of user-generated content on social media platforms has enabled new opportunities for computational emotion and threat analysis using machine learning (ML) techniques. This paper presents a comprehensive systematic review of recent advancements in detecting emotional states and threats from social media data, covering literature published between 2019 and 2025. The review follows PRISMA guidelines and includes ten peer reviewed studies selected from an initial pool of 162 records across databases such as IEEE Xplore, Google Scholar, Elsevier, and SpringerLink. The included works are categorized into three domains: emotion detection, threat and crisis detection, and hybrid emotion-aware applications such as hate speech classification and social signal processing. The analysis reveals that deep learning models, particularly CNN, Bi-LSTM, GRU, and transformer-based architectures like BERT, significantly outperform traditional methods in emotion classification and threat recognition tasks. Several works have focused on leveraging machine learning (ML) and deep learning (DL) models to detect emotional expressions in online content. A real time emotion detection framework Smart Mood Detection using deep neural networks to enhance human computer interaction. Integration of emotion features enhances detection of cyber threats and hate speech, while optimization algorithms further improve accuracy and generalizability. Key challenges identified include dataset imbalance, lack of multilingual resources, limited explainability, and real-time deployment barriers. This review synthesizes current methodologies, highlights gaps, and offers a research agenda for building intelligent, emotion-aware systems for social media-based threat detection.