Abstract
Textual emotion detection is a critical area of study with significant applications in business, education, and healthcare. Despite substantial theoretical advancements over the years, there is a notable gap in the practical implementation of these methods in the aforementioned fields. The techniques currently available do not yet seem ready for real-world application. This study offers a comprehensive review of existing approaches, datasets, and models used in textual emotion detection. Its primary objective is to identify the challenges faced in both current literature and practical applications. The findings reveal that textual datasets annotated with emotional markers are scarce, making it difficult to develop robust supervised classification models for this task. There is also a pressing need for improved models that can accurately categorize a wider range of emotional states distinctly. Finally, there is a demand for techniques capable of dimensionally detecting valence, arousal, and dominance scores from emotional experiences. These challenges stem not only from the models and applications themselves but also from the readiness of current approaches and datasets in the rapidly evolving fields of machine learning and affective computing.