Enhancing Agriculture QA Models Using Large Language Models

Mathematics, University of California San Diego, La Jolla, USA

^* Corresponding author: x6he@ucsd.edu

Abstract

Agriculture is a complex process that requires a great deal of knowledge and experience. Yet, the complexity and often chaotic nature of web data presents a considerable challenge in obtaining suitable results. Given these production requirements, there is an urgent need to develop a model that enables machine reading comprehension and caters to automatic question-answering scenarios within the scope of agricultural production. In this paper, we construct a dataset for the experiments of document QA in agricultural scenarios. We import two model (ask-my-pdf and chat-pdf) as our baseline and use them to do the single and multiple document QA task. Then, we proposed several methods to improve the performance of the model in agriculture scenarios. At the end of the experiment, we achieved 33.3% improvement in F1 score compared to baseline and 92.8% overall answer accuracy in single document QA. For our final multiple documents QA model, we achieved 53% improvement in F1 score compared to baseline and 83.3% overall answer accuracy in the task.