BUILDING AN ADMISSION CONSULTING CHATBOT FOR CAN THO UNIVERSITY

Abstract

The growing demand for fast and accurate information in university admissions requires smart and collaborative solutions. This thesis presents the design and implementation of a chatbot system to assist with admission counseling at Can Tho University (CTU). The proposed chatbot uses recent advances in natural language processing (NLP) and artificial intelligence, combining optimized macrolinguistic models (LLMs) with Retrieval-Augmented Generation (RAG) techniques, vector databases, and knowledge graphs to provide accurate and contextual answers. The system uses the optimized LLMA3-8B LLM built on state-of-the-art architectures, specifically adapted to understand and generate human-like responses in the field of CTU receptions. To improve the accuracy of information retrieval, a vector store-Qdrant is used to identify and retrieve relevant data elements, which are extracted and inserted from the comprehensive corpus of admission-related documents. The incorporation of the RAG framework allows the chatbot to combine acquired factual data with generative language capabilities, ensuring the delivery of reliable and informative answers. In addition, the knowledge graph is integrated to represent structured relationships between admission-related entities, enabling more complete and explicable answers. Experimental results show that the chatbot effectively addresses common acceptance queries with high relevance and user satisfaction, outperforming traditional acquisition or rule-based processes. The application of fine-tuning and RAG techniques greatly enhances chatbot understanding and response quality. This project provides a scalable and flexible framework that can be used by other educational institutions looking to enhance their admissions counseling services with AI-driven chat agents.