Emotion Classification Using Transformer-Based Language Model

Abstract

This paper presents Emotion Classification Using Transformer-Based Language Models, highlighting the growing interest in emotion-related studies, particularly in the Thai language. Emotions play a crucial role in perception, influencing text input on various platforms or responses in chatbot interactions. Text messages, when combined into sentences, may convey diverse emotions, leading to misunderstandings and potentially inappropriate behavior. To classify emotions in text, experiments were conducted using a dataset from the Jubjai chatbot. The study aimed to identify the most effective pre-trained model and compare the results of data cleansing between fine-tuned and non-fine-tuned models to evaluate the accuracy in analyzing 7, 5, and 3 emotions. The experimental results demonstrated that the fine-tuned Wangchangberta model outperformed XLM-RoBERTa, achieving accuracy rates of 73% for 7 emotions, 76% for 5 emotions, and 82% for 3 emotions.