Editorial: Semantics and Natural Language Processing in Agriculture

Mathieu Roche¹Brett Drury^2*

• ¹Centre de coopération Internationale en Recherche Agronomique pour le Développement, Paris, France
• ²Liverpool Hope University, Liverpool, United Kingdom

The application of data science to agriculture enables the analysis of diverse agricultural data through methods such as machine learning, deep learning, computer vision, text mining [5], and large language models (LLMs). These approaches support tasks including crop yield prediction and the early detection of plant and animal diseases by exploiting heterogeneous data sources (e.g., sensor data, textual reports, satellite imagery, and plant images). For practitioners and decision makers, data-driven insights provide a sound evidence base for more efficient and sustainable agricultural practices.Within digital agriculture, textual data and semantic information remain central challenges. Semantics assigns concepts a unique identifier, reducing ambiguity and facilitating data integration across the agricultural value chain. This improves interoperability and strengthens agricultural information systems. In this context, Agrisemantics has emerged as a dedicated approach to semantic technologies in agriculture [4].The urgency of this research is reinforced by global pressures, including rising food prices, reduced arable land linked to climate change, and population growth. This Research Topic attracted four contributions, comprising original research, brief reports, and data studies, all highlighting advances in semantics and text mining for agricultural applications. Chaib, Macombe, and Thomopoulos [2] proposed a systematic methodology for developing agricultural ontologies by combining the Godet and MyChoice approaches. Starting from stakeholder interviews and extending to complementary construction techniques, their work illustrates the benefits of integrating methodologies to strengthen ontology design in agriculture. Food safety remains a global concern, underscored by incidents such as Operac ¸ão Carne Fraca in Brazil and the 2008 Chinese Milk Scandal. Yet even uncontaminated food products may generate perceptions of risk among consumers. Aline, Hubert, Pitarch, and Thomopoulos [1] examined subjective consumer beliefs regarding the risks of infant food through a cross-national study. Their analysis revealed intracultural variations in risk perception, offering insights that can inform governmental communication strategies aimed at alleviating public concerns. In addressing data sparsity in agricultural text classification, Jiang, Cormier, Angarita, and Rousseaux [6] employed unlabelled data and Generative Adversarial Networks to fine-tune a pre-trained language model (PLM). Their method demonstrated improved performance over traditional approaches across multiple tasks. By reducing reliance on costly labelled datasets, their approach lowers barriers to adopting text mining for agricultural applications. Rakotomalala et al. [8] developed domain-specific lexicon to support research on organic residue valorisation in developing countries. Combining expert knowledge with NLP techniques, the authors identified 2,079 relevant terms, made publicly available (https://doi.org/10.18167/DVN1/ HNZZSI). The lexicon has since enabled semantic-driven analyses of a broad corpus compiled from multiple scientific databases and repositories. Text mining and semantic approaches are becoming increasingly prominent in agricultural research. Since the launch of this Research Topic, contributions have collectively attracted more than 11,000 views. These efforts also pave the way for future studies applying LLMs to agricultural challenges [3,7]. The support of organisations such as GODAN ( Global Open Data for Agriculture and Nutrition) 1 and the FAOfoot_1 will be crucial for advancing applications that integrate semantics and text mining at the forefront of agricultural research and practice.