DSQA-LLM: Domain-Specific Intelligent Question Answering Based on Large Language Model

Question Answering (QA) is crucial for humans to access vast knowledge bases, but there is a lack of attention towards representing raw, unstructured questions and answers in specific fields. Additionally, the efficiency of finding candidate questions based on the trigger question and the generation of reasonable answers have been neglected. In this paper, we introduce Domain Specific Question Answering Language Model (DSQA-LLM), a framework that delivers informative answers within a specific domain. We utilize techniques like question classification, information retrieval, and answer generation. We enhance efficiency and accuracy through the integration of XLNET for question classification and a novel similarity searching method using Sentence-T5. Furthermore, the powerful GPT-3.5-turbo is employed for generating coherent answers. We implemented DSQA-LLM and curated a dataset of 127,840 question-answer pairs. Empirical experiments conducted on real-world questions confirm the effectiveness of our QA system.