Improving Thai Sentiment Analysis Accuracy with Emoji Classification by Deep Learning and Stacking Models: A Case Study of Hotel Reviews
Walaithip Bunyatisai, Suwika Plubin, Knavoot Jiamwattanapong, Bandhita Plubin · Pakistan Journal of Life and Social Sciences (PJLSS) · 2024
Sentiment analysis presents unique challenges in hotel reviews, particularly in languages like Thai, renowned for nuanced expressions.Understanding customer opinions is pivotal, especially in domains such as hotel reviews, where subjective expressions prevail.This study delves into methodologies to refine sentiment analysis accuracy in Thai hotel reviews by integrating emoji classification with Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and stacking models.Through comparing models with and without emoji inclusion, we meticulously evaluate their performance metrics encompassing accuracy, precision, recall, and F1-score.The pinnacle achievement of 92.4% accuracy underscores the efficacy of advanced stacking techniques complemented by emoji integration.Our findings underscore the superiority of models encompassing emojis, affirming the value of amalgamating textual and emoji data.Leveraging sophisticated deep learning techniques and stacking models, our approach adeptly captures the subtle nuances of sentiment expressed in Thai text, resulting in heightened accuracy in sentiment analysis.This research underscores the paramount importance of embracing diverse data sources and sophisticated modeling strategies to elevate sentiment analysis accuracy in Thai hotel reviews.