Abstract

Hate speech is a form of communicative expression that promotes or incites unjustified violence. The increase in hate speech on social media has prompted the development of automated tools for its detection, especially those that integrate emotional tone analysis. This study presents a systematic review of the literature, employing a combination of PRISMA and PICOS methodologies to identify the most used Natural Language Processing and Machine Learning techniques for emotion classification in hostile messages. It also seeks to determine which models and tools predominate in the analyzed studies. The findings highlight LLaMA 2 and HingRoBERTa, achieving F1 scores of 100% and 98.45%, respectively. Furthermore, key challenges are identified, including linguistic bias, language ambiguity, and the high computational demands of some models. This review contributes an updated overview of the state of the art, highlighting the need for more inclusive, efficient, and interpretable approaches to improve automated moderation on digital platforms. Additionally, include techniques, methods, and future directions in this topic.