vetpressАграрная наукаAgrarian science0869-81552686-701XРедакция журнала "Аграрная наука"10.32634/0869-8155-2025-395-06-162-166vetpress-3707Research ArticleЦИФРОВИЗАЦИЯ АПКDIGITALIZATION OF THE AGRO-INDUSTRIAL COMPLEXЭффективные методы адаптации LLM к доменной специфике аграрного бизнесаEffective Methods for Adapting LLM to the Agricultural Business DomainКапитановА. И.KapitanovA. I.

Андрей Иванович Капитанов, кандидат технических наук, доцент

пл. Шокина, 1, Москва, 124498

Andrey Ivanovich Kapitanov, Candidate of Technical Sciences, Associate Professor

1 Shokina Square, Moscow, 124498

andrey@kapdx.ruНациональный исследовательский университет «МИЭТ»РоссияNational Research University “MIET”Russian Federation20252406202506162166Copyright © Капитанов А.И., 20252025Капитанов А.И.Kapitanov A.I.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://www.vetpress.ru/jour/article/view/3707

В статье рассматриваются актуальные проблемы применения больших языковых моделей (LLM) в сфере аграрного бизнеса и предлагаются современные подходы к их решению. Несмотря на высокую эффективность LLM в обработке естественного языка, их адаптация к задачам аграрной отрасли связана с рядом сложностей. Ключевые проблемы включают формирование специализированных обучающих корпусов, баланс между качеством ответов и вычислительными затратами, объективную оценку качества моделей и их интеграцию в существующие аграрные информационные системы. Рассматриваются практические подходы к решению этих проблем, включая дообучение моделей на специализированных данных, методы оптимизации вычислений и применение гибридных архитектур (в частности, RAG). Анализируются основные направления применения LLM: генерация текстовых данных, улучшение поисковых систем, анализ пользовательских отзывов и автоматизация клиентской поддержки. Исследование направлено на повышение точности, релевантности и персонализации ответов моделей в задачах прогноза, анализа и автоматизации процессов в сельском хозяйстве. Предложенные решения способствуют эффективному внедрению LLM в инфраструктуру аграрного сектора, улучшая качество принятия решений, прогнозирование и автоматизацию бизнес-процессов.

The article discusses the current challenges of applying large language models (LLMs) in the agricultural business sector and proposes modern approaches to address these issues. Despite the high effectiveness of LLMs in natural language processing, their adaptation to the tasks of the agricultural industry involves a number of difficulties. Key problems include the formation of specialized training corpora, balancing the quality of responses with computational costs, objective evaluation of model quality, and their integration into existing agricultural information systems. Practical approaches to solving these problems are discussed, including fine-tuning models on specialized data, computational optimization methods, and the use of hybrid architectures (in particular, RAG). The main areas of LLM application are also analyzed: text generation, search engine improvement, analysis of user reviews, and customer support automation. The research aims to improve the accuracy, relevance, and personalization of model responses in tasks related to forecasting, analysis, and automation of processes in agriculture. The proposed solutions contribute to the effective integration of LLMs into the infrastructure of the agricultural sector, enhancing decision-making quality, forecasting, and business process automation.

LLMfine-tuningRAGсемантический поискадаптация моделиперсонализациягенеративные моделиLLMfine-tuningRAGsemantic searchmodel adaptationpersonalizationgenerative modelsReferencesPan S., Luo L., Wang Y., Chen C., Wang J., Wu X. Unifying Large Language Models and Knowledge Graphs: A Roadmap. IEEE Transactions on Knowledge and Data Engineering. 2024; 36(7): 3580–3599. https://doi.org/10.1109/TKDE.2024.3352100Pan S., Luo L., Wang Y., Chen C., Wang J., Wu X. Unifying Large Language Models and Knowledge Graphs: A Roadmap. IEEE Transactions on Knowledge and Data Engineering. 2024; 36(7): 3580–3599. https://doi.org/10.1109/TKDE.2024.3352100Hu E.J. et al. LoRA: Low-Rank Adaptation of Large Language Models. arXiv. 2106.09685. https://doi.org/10.48550/arXiv.2106.09685Hu E.J. et al. LoRA: Low-Rank Adaptation of Large Language Models. arXiv. 2106.09685. https://doi.org/10.48550/arXiv.2106.09685Feldman P., Dant A., Foulds J.R., Pan S. Polling Latent Opinions: A Method for Computational Sociolinguistics Using Transformer Language Models. arXiv. 2204.07483. https://doi.org/10.48550/arXiv.2204.07483Feldman P., Dant A., Foulds J.R., Pan S. Polling Latent Opinions: A Method for Computational Sociolinguistics Using Transformer Language Models. arXiv. 2204.07483. https://doi.org/10.48550/arXiv.2204.07483Rusnachenko N., Golubev A., Loukachevitch N. Large Language Models in Targeted Sentiment Analysis for Russian. Lobachevskii Journal of Mathematics. 2024; 45(7): 3148–3158. https://doi.org/10.1134/S1995080224603758Rusnachenko N., Golubev A., Loukachevitch N. Large Language Models in Targeted Sentiment Analysis for Russian. Lobachevskii Journal of Mathematics. 2024; 45(7): 3148–3158. https://doi.org/10.1134/S1995080224603758Krishnan A. Exploring the Power of Topic Modeling Techniques in Analyzing Customer Reviews: A Comparative Analysis. arXiv. 2308.11520. https://doi.org/10.48550/arXiv.2308.11520Krishnan A. Exploring the Power of Topic Modeling Techniques in Analyzing Customer Reviews: A Comparative Analysis. arXiv. 2308.11520. https://doi.org/10.48550/arXiv.2308.11520Lewis P. et al. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. arXiv. 2005.11401. https://doi.org/10.48550/arXiv.2005.11401Lewis P. et al. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. arXiv. 2005.11401. https://doi.org/10.48550/arXiv.2005.11401Chamieh I., Zesch T., Giebermann K. LLMs in Short Answer Scoring: Limitations and Promise of Zero-Shot and Few-Shot Approaches. Proceedings of the 19th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2024). Association for Computational Linguistics. 2024; 309–315.Chamieh I., Zesch T., Giebermann K. LLMs in Short Answer Scoring: Limitations and Promise of Zero-Shot and Few-Shot Approaches. Proceedings of the 19th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2024). Association for Computational Linguistics. 2024; 309–315.Chen X. et al. Universal Self-Consistency for Large Language Model Generation. arXiv. 2311.17311. https://doi.org/10.48550/arXiv.2311.17311Chen X. et al. Universal Self-Consistency for Large Language Model Generation. arXiv. 2311.17311. https://doi.org/10.48550/arXiv.2311.17311Lin C.-Y. ROUGE: A Package for Automatic Evaluation of Summaries. Text Summarization Branches Out. Association for Computational Linguistics. 2004; 74–81.Lin C.-Y. ROUGE: A Package for Automatic Evaluation of Summaries. Text Summarization Branches Out. Association for Computational Linguistics. 2004; 74–81.Zhang T., Kishore V., Wu F., Weinberger K.Q., Artzi Y. BERTScore: Evaluating Text Generation with BERT. arXiv. 1904.09675. https://doi.org/10.48550/arXiv.1904.09675Zhang T., Kishore V., Wu F., Weinberger K.Q., Artzi Y. BERTScore: Evaluating Text Generation with BERT. arXiv. 1904.09675. https://doi.org/10.48550/arXiv.1904.09675Banerjee S., Lavie A. METEOR: An Automatic Metric for MT Evaluation with Improved Correlation with Human Judgments. Proceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and (or) Summarization. Association for Computational Linguistics. 2005; 65–72.Banerjee S., Lavie A. METEOR: An Automatic Metric for MT Evaluation with Improved Correlation with Human Judgments. Proceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and (or) Summarization. Association for Computational Linguistics. 2005; 65–72.

The authors declare that there are no conflicts of interest present.