Nuttawut Thuayhanruksa. Comparative study of llm models for sentiment classification in Thai financial news headlines. Master's degree(Data Science). Chiang Mai University. Library. : Chiang Mai University, 2025.
Comparative study of llm models for sentiment classification in Thai financial news headlines
Abstract:
This research explores the application of various natural language processing (NLP) models for sentiment analysis on financial news headlines from the Kaohoon, Hoonsmart, and Investing websites, with a specific focus on Thai language data. The study evaluates the performance of several machine learning and deep learning models, including Logistic Regression, Bidirectional Long Short-Term Memory (Bi-LSTM), Convolutional Neural Networks (CNN), OpenAIs GPT-3.5, WangChanBERTa, and OpenThaiGPT. The research employs a train-test split with a fixed random state for complex models and stratified cross-validation for basic models, ensuring reproducibility and consistency in results. Model performance is assessed using evaluation metrics such as accuracy, precision, recall, and F1-score. The findings demonstrate the effectiveness of these models in analyzing sentiment in financial news, providing insights for improving sentiment analysis in Thai financial contexts. The study's results reveal that the Fine-tuned WangChanBERTa model outperformed others in sentiment analysis, achieving the highest accuracy of 0.84 on the testing set, demonstrating its strong ability to classify sentiment in Thai financial news headlines. The BI-LSTM and CNN models also showed competitive performance, with testing accuracies of 0.781 and 0.791, respectively. In contrast OpenAIs GPT-3.5 and OpenThaiGPT models did not include fine-tuned and optimize prompt due to the high computational cost and time required for fine-tuning, this highlights the practical limitations of using large-scale models in resource-constrained scenarios.