Luận văn Thạc sĩ Công nghệ thông tin: Xây dựng Chatbot bán hàng dựa trên mô hình sinh

The burgeoning landscape of social media and e-commerce has amplified the demand for efficient and responsive customer interaction. Traditional manual support systems often prove costly and inefficient in handling high volumes of repetitive inquiries, leading to missed opportunities and decreased customer satisfaction. This context highlights the critical need for intelligent, automated solutions. This thesis addresses this imperative by focusing on the development of a sales chatbot, leveraging generative models to enhance interaction quality. The primary objective is to investigate and apply advanced natural language processing (NLP) techniques for robust intent classification, information extraction, and dynamic dialogue management, ultimately building a sophisticated chatbot capable of providing natural and effective support within a closed-domain sales environment.

E-commerce businesses, customer service managers, natural language processing researchers, AI developers, and students interested in chatbot technology and its application in sales.

This master's thesis presents a comprehensive approach to building an intelligent sales chatbot, addressing the escalating need for automated customer support in the digital commerce era. The study begins by analyzing the limitations of conventional chatbot models, such as script-based and keyword-recognition systems, advocating for a more sophisticated generative model approach. It delves into the architectural components of a modern chatbot, meticulously detailing Natural Language Understanding (NLU) for precise intent classification and entity extraction, Dialogue Management (DM) for maintaining conversational context and decision-making, and Natural Language Generation (NLG) for crafting human-like responses. The methodology incorporates cutting-edge artificial intelligence techniques, including various Neural Networks (RNN, LSTM) for handling sequential data, Conditional Random Fields (CRF) for information extraction, and advanced word embedding methods like Word2vec and GloVe. Notably, the study utilizes the open-source Rasa framework for implementation, focusing on its adaptability for Vietnamese language processing and its ability to manage complex dialogues in a closed-domain sales context. Practical aspects, such as data collection and training using real-world Facebook Fanpage interactions, are emphasized, yielding valuable insights into data labeling and model refinement for improved accuracy. The resulting chatbot significantly enhances sales efficiency, reduces operational costs, and fosters more natural and satisfactory customer engagements, demonstrating the tangible benefits and high practical applicability of integrating generative AI into e-commerce customer service.