Research on Cucumber Knowledge QA Systems with Integrated Large Language Models

Traditional question answering (QA) methods often lack flexibility, and general-purpose large language models (LLMs) are prone to factual errors when generating answers in the agricultural domain. To address these issues, this paper proposes CuQA, a cucumber-domain QA approach that integrates improved alignment, intent recognition, multi-channel retrieval, re-ranking, and prompt engineering. By effectively combining these components, a high-quality cucumber knowledge base is built to support accurate and domain-specific answer generation using LLMs. Experimental results show that the proposed method significantly improves knowledge retrieval relevance. The generated answers outperform those of the baseline and other general models by more than 15.2 % on metrics such as METEOR and ROUGE-L. Evaluations conducted using DeepSeek also show that CuQA consistently outperforms comparative models. The study confirms that integrating external domain knowledge with LLMs can effectively improve the accuracy of question answering in the cucumber domain.