Comparative Analysis of Traditional and Large Language Model Techniques For Multi-Class Emotion Detection
Authors
Kuppachi, Madhumitha
Issue Date
2024
Degree
MSc in Artificial Intelligence
Publisher
Dublin Business School
Rights holder
Rights
Abstract
In recent years, YouTube's comment sections have drawn interest in analyzing the emotions within them. This study investigates the emotion detection of YouTube comments, seeking to improve accuracy by testing machine learning and deep learning models. We compare traditional models - Logistic Regression, Naive Bayes, Decision Trees, Support Vector Machines, Random Forests, and newer deep learning models like - LSTMs, BERT, and DistilBERT. Recent deep learning advancements have shown promise for emotion detection. We aim to determine the most effective approaches.Although traditional models offered insightful categorizations, deep learning structures like LSTM and DistilBERT displayed remarkable capabilities, indicating their potential for very accurate emotion detection. These results give key understandings of choosing models based on specific needs, highlighting the necessity for ongoing improvement and optimization, especially in deep learning methods, to reach greater precision and durability in the classification of emotions within YouTube video comments.