AgriPrompt: A Method to Enhance ChatGPT for Agricultural Question Answering
Tianyue Chen, Xiaojin Chen, Yongqiang Qian, Lang Zheng, Haiyang Li, Jingbo Zhao, Yaojun Wang · 2024
We propose a method called AgriPrompt to enhance the agricultural question-answering capability of ChatGPT. We propose a BERT-based model, AgriParse, to extract semantic information from agricultural questions, which is filled into pre-defined prompt templates to guide ChatGPT in answering agricultural questions. We establish an evaluation metric called KQScore, which uses the TF-IDF method and the Word2vec model to evaluate the semantic accuracy of answers. The results show that the average KQScore of ChatGPT with AgriPrompt is 0.81, ChatGPT without prompt is 0.74, and ChatGPT with a static prompt is 0.72. Our method significantly improves the accuracy of ChatGPT in answering agricultural questions.