Abstract: Emotion detection in text is the task of determining the feelings that appear when writing or reading a text. Emotion detection has been addressed as a multi-label problem, as different emotions can be present simultaneously in a given text. In the past few years, many multi-label methods have been developed with different philosophies and objectives. However, a comparison of all the many possible multi-label classification models is still lacking. Thus, when facing an emotion detection problem from a multi-label point of view, many available methods are at the disposal of the researcher without a clear knowledge of which is best. In this paper, we present a thorough comparison using a large set of 29 multi-label state-of-the-art classification models and 18 datasets of real-world emotion detection problems. We test both single classifiers and ensemble methods. Our analysis shows that multi-label methods achieve competitive performance. ML-DT, IBLR-ML+, and MLARAM show the best overall performance considering single classifiers. RAkEL-O and ECC achieve the best results for ensemble methods.
J. M. Cuevas-Muñoz, N. García-Pedrajas, and A. de Haro-García (2025) “Comprehensive comparison of multi-label methods for emotion detection,” submitted.
Supplementary material: