Emotion Detection in Conflict Discourse: An Unsupervised Approach to Multilingual YouTube Comments

Abstract

Introduction: Emotion detection from text has become a critical research direction in natural language processing, particularly with the rise of emotionally charged and multilingual content on platforms like YouTube.

Objectives: The main objectives of this study are the automatic annotation of a multilingual corpus and the investigation of the psychological impact of crises on individuals, by analyzing the emotional content embedded in YouTube comments associated with war-related videos.

Methods: This study focuses on unsupervised emotion detection from YouTube comments related to the Gaza war, using data in both English and Arabic. It explores the role of Emojis in enhancing emotional interpretation. Generally, the findings confirm the effectiveness of the clustering techniques in detecting emotional patterns from unstructured and diverse YouTube data. The proposed framework presents a scalable, language-aware solution for unsupervised emotion detection, making it particularly suitable for multilingual and sensitive contexts.  

Results: The resulting labeled datasets provide a valuable foundation for training supervised models, thereby contributing to the development of more robust and effective emotion detection systems in the domain of YouTube comment analysis. The results highlights a consistent predominance of negative emotions across both English and Arabic datasets, regardless of emoji inclusion.

Conclusions: Negative emotions exceed 70% in the English data and surpass 80% in the Arabic data, while positive emotions remain notably underrepresented in all cases.