Campus question-answering system based on intent recognition and retrieval-augmented generation

Abstract

To address the issues of poor information integration and generalization in traditional campus question-answering systems, a campus question-answering system based on a large language model was designed. The fine-tuned model identified user intents and provided targeted solutions for various types of questions, enhancing the user experience. To mitigate the hallucination problem during language model generation, a knowledge base using diverse campus data was constructed and a retrieval-augmented generation method was employed to ensure factual accuracy. Experimental results indicate that the open-source large language model, after instruction tuning, achieves intent recognition accuracy that is comparable to or even surpasses that of closed-source models.

Keywords

• LLM
• RAG
• supervised fine-tuning
• intent recognition