Fine-Grained Classification for Emotion Detection Using Advanced Neural Models and GoEmotions Dataset

Abstract

Emotion detection, a pivotal facet of artificial intelligence, involves deciphering and categorizing human emotions from various sources such as text, images, and audio. This process holds immense significance across industries, including mental health, customer sentiment analysis, and human-computer interaction. This abstract encompasses the essence of emotion detection, its vital role in understanding human behaviors and sentiments, and the diverse methods employed for this purpose. The study comprises an exploration of three distinct emotion detection techniques: ROBERTA, BERT, and CNN. These methods are evaluated for their effectiveness in recognizing a range of emotions, from subtle nuances to more overt expressions. The results reveal ROBERTA's exceptional prowess, consistently outperforming its counterparts across various emotional categories. Particularly remarkable is its ability to predict the emotion "gratitude" with an impressive F1-score of 0.8458, underscoring its potential in capturing complex emotional states. This research emphasizes the significance of emotion detection in bridging human-computer interaction gaps and enabling more nuanced understanding of user sentiments. The findings emphasize prominence of ROBERTA as a powerful tool in emotion detection, offering insights into its capacity to comprehend diverse human emotions effectively.