Abstract

One of the persistent challenges in agricultural statistics is the lack of standardized frameworks for categorizing emerging technologies, hindering cross-national comparisons and evidence-based policy development in the agricultural sector. This study presents a novel artificial intelligence-driven approach to systematically categorize and analyze agrifood technology patents, creating a comprehensive dual-taxonomy framework that supports evidence-based policy development and technology assessment.

Employing more than 3.5 million agrifood systems patent applications from 1980 until 2025 using PATSTAT, we develop a two-layered taxonomical framework that simultaneously classifies agricultural patents by (1) underlying technology characteristics and (2) intended purpose or application domain. Using a fine-tuned large language model (LLM), we extract structured information from patent titles and abstracts, capturing both technical specifications and functional objectives. The extracted data undergoes a data-driven clustering approach, generating hierarchical categorizations at macro (broad technological domains), meso (specific technology clusters), and micro (granular technology applications) levels for both dimensions.

Unlike traditional patent classification systems that rely on predetermined categories, our approach employs unsupervised machine learning to discover emergent patterns in the data, revealing previously unrecognized connections between technologies and applications. This dual-taxonomy structure enables multidimensional analysis, allowing users to explore patents by technical approach (e.g., sensor technologies, biotechnology applications) or by agricultural purpose (e.g., precision farming, sustainable production systems).

The resulting taxonomy serves as the foundational framework for the FAO's ATIO (Agricultural Technology Intelligence Observatory) Knowledge base, providing agricultural statisticians, policymakers, and researchers with a systematic tool for technology landscape analysis. This enables evidence-based assessment of innovation trends, identification of technology gaps, and informed policy development for agricultural transformation.

The framework addresses critical needs in agricultural statistics by providing: (1) standardized categorization of emerging technologies for cross-national comparisons; (2) systematic tracking of innovation patterns to inform research and development priorities; (3) identification of technology clusters relevant to sustainable development goals; and (4) enhanced capacity for evidence-based policy formulation in the digital agriculture domain.

This work contributes to the growing field of agricultural data science by demonstrating how advanced AI techniques can transform patent data into actionable intelligence for agricultural policy and innovation management. The approach offers a scalable, adaptable framework that can evolve with emerging technologies while maintaining consistency in categorization standards essential for longitudinal agricultural statistics and policy analysis.