Tóm tắt Đề án tốt nghiệp Thạc sĩ: Ứng dụng học tăng cường trong phân tích cảm xúc theo chủ đề

The rapid growth of digital transformation and user-generated content, especially online customer feedback, presents an invaluable resource for businesses in Vietnam. However, extracting meaningful insights from this vast, unstructured, and emotionally nuanced data, particularly in the complex Vietnamese language, poses significant challenges for traditional analysis methods. This thesis highlights the critical importance of leveraging advanced artificial intelligence techniques, notably Reinforcement Learning, to overcome these limitations. By adopting a trial-and-error mechanism with continuous environmental interaction, học tăng cường offers a dynamic approach to analyze intricate linguistic patterns. This research aims to develop a robust solution for phân tích cảm xúc theo chủ đề, thereby enabling businesses to precisely understand customer sentiment across specific product or service aspects and enhancing their competitive advantage.

Researchers and practitioners in artificial intelligence, machine learning, natural language processing, data science, and businesses focused on customer experience, market research, and product development, especially those dealing with Vietnamese language data.

This research addresses the complex challenge of phân tích cảm xúc theo chủ đề (Aspect-Based Sentiment Analysis – ABSA) within online customer feedback, particularly for the Vietnamese language. Traditional approaches, including rule-based methods, conventional Machine Learning, and even deep learning models like LSTM or BERT, often struggle with the unstructured, diverse, and nuanced nature of user-generated content, especially concerning the intricacies of tiếng Việt. To overcome these limitations, the thesis proposes a novel integration of Reinforcement Learning (RL) into the ABSA framework. RL, as a branch of trí tuệ nhân tạo, offers a dynamic learning paradigm where an agent interacts with the environment, learns from feedback (rewards), and continuously optimizes its decision-making policy. The methodology involves exploring fundamental RL algorithms such as Q-Learning, Deep Q-Network (DQN), and Actor-Critic, alongside advanced xử lý ngôn ngữ tự nhiên (NLP) techniques like BERT for feature extraction. The study focuses on analyzing unstructured ý kiến khách hàng data, encompassing both Vietnamese and English, to classify sentiments (positive, negative, neutral) and identify specific aspects like product quality, service, and pricing. This approach promises enhanced accuracy and adaptability compared to static models, facilitating continuous learning and performance improvement through experience. The practical implications are significant for e-commerce and customer service, offering businesses a more precise understanding of customer needs and satisfaction, ultimately leading to improved products, services, and strategic marketing decisions.