AGRI-Check: Fact-Checking Agricultural Misinformation

Rakesh Thakur, Halija Vahaj, Vriddhi Agarwal

Keywords: Agricultural Misinformation, Fact-Checking, Benchmark Dataset, Claim Verification, Evidence Retrieval, Large Language Models (LLMs), Quantum-enhanced LLMs, Natural Language Processing (NLP), Justification Generation, Domain-specific Misinformation Detection

TL;DR: AGRI-Check is the first agriculture-focused fact-checking dataset, benchmarking classical and quantum LLMs for detecting and correcting misinformation on crops, fertilizers, policies, and weather.

Abstract: Agricultural misinformation spanning fertilizers, pesticides, crop diseases, government policies, and weather forecasts poses serious risks to farmer decision-making and food security. Despite advances in automated fact-checking for politics and healthcare, agriculture remains largely overlooked. To address this gap, we introduce AGRI-Check, a benchmark dataset and framework for agricultural misinformation detection. Each entry in AGRI-Check is structured by category and includes the original claim, a misinformation variant, and the correct authoritative information from trusted sources such as ICAR, FAO, IMD, and official government documents. This design supports both misinformation detection and correction. We further conduct a comparative study of conventional large language models (LLMs) and quantum-enhanced LLMs for claim verification, integrating evidence retrieval, classification, and justification generation. Results show that while classical LLMs perform strongly in textual reasoning, quantum-based models offer efficiency gains and improved robustness for ambiguous cases. AGRI-Check establishes the first domain-specific benchmark in agriculture and advances quantum NLP applications in fact-checking.

Primary Area: generative models

Submission Number: 23994