pdKGraph: A novel approach to constructing plant disease knowledge graphs using large language models

Authors: MIRTA, ALAKANANDA - University Of Nebraska-Lincoln; DUONG, DINH QUY - University Of Nebraska-Lincoln; Fleisher, David; Reddy, Vangimalla; RAY, CHITTARANJAN - University Of Nebraska-Lincoln

Submitted to: IEEE Access
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/12/2026
Publication Date: N/A
Citation: N/A

Interpretive Summary: Plant diseases continue to threaten U.S. and global agriculture, causing economic losses and undermining food security. However, current tools for early detection and knowledge synthesis are limited, leaving researchers struggling to keep pace with evolving pathogens and diagnostic needs. As an initial step toward addressing this gap, we developed an artificial intelligence powered framework. Our system organized information on diseases, detection methods, datasets, and performance metrics into a knowledge graph which substantially improves the ease of obtaining disease detection information at the farm location. The system was shown to be highly accurate and easy to use by end-users. This early-stage research lays the groundwork for future tools that can support more adaptive and insight-driven approaches in plant disease diagnostics and agricultural research and benefits agronomists, crop extension agents, farm managers and researchers.

Technical Abstract: Plant diseases represent a persistent challenge to global agriculture, resulting in significant economic losses and posing a threat to food security. Traditional diagnostic approaches and static literature reviews are increasingly inadequate in keeping pace with rapidly evolving pathogens and emerging detection technologies. In this initial study, we present a prototype artificial intelligence framework that integrates Retrieval-Augmented Generation (RAG) with Knowledge Graphs (KGs) to support intelligent exploration and structuring of plant disease literature. The system enables semantic retrieval, natural language querying, and automated extraction of key entities—including diseases, diagnostic methods, datasets, and performance metrics—into a structured KG. Achieving a Top-1 accuracy of 70.3 percent, with an average query latency of 13–16 seconds and a hallucination rate of 10.8 percent, the framework demonstrates the feasibility of combining RAG and KGs for dynamic, evidence-grounded information access. This foundational work establishes a scalable basis for future research into more advanced reasoning capabilities, including temporal trend analysis and knowledge gap identification in the domain of plant disease diagnostics.