Scholarlink

Knowledge retrieval plays an essential role in decision-making, and it becomes a challenge for farmers due to the complexity of farming data. In this work, an ensemble model that integrates general-purpose and domain-specific LLMs (BERT-base uncased, Agricultural-BERT and LLaMA 3.1) to enhance response correctness and appropriateness for agricultural questions. Using domain-specific data and a weighted voting scheme, our ensemble model demonstrates significant effectiveness in improving response accuracy, achieving approximately 94.6% accuracy, a BLEU score of 54.7, and a ROUGE-1 score of 0.73, as confirmed by performance evaluation. Results demonstrate that this system can assist farmers by providing actionable insights, especially for agricultural management and crop health. This ensemble model represents a better solution for advancing agriculture knowledge retrieval by integrating powerful but general LLM with domain-specific knowledge to tackle its unique challenges.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Weighted Ensemble Approach Integrating Large Language Models for Enhanced Agricultural Knowledge Retrieval

  • 会议论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年

查看全文
  • Cyreneo Dofitas Jr,
  • Yong-Woon Kim,
  • Yung-Cheol Byun

摘要

Knowledge retrieval plays an essential role in decision-making, and it becomes a challenge for farmers due to the complexity of farming data. In this work, an ensemble model that integrates general-purpose and domain-specific LLMs (BERT-base uncased, Agricultural-BERT and LLaMA 3.1) to enhance response correctness and appropriateness for agricultural questions. Using domain-specific data and a weighted voting scheme, our ensemble model demonstrates significant effectiveness in improving response accuracy, achieving approximately 94.6% accuracy, a BLEU score of 54.7, and a ROUGE-1 score of 0.73, as confirmed by performance evaluation. Results demonstrate that this system can assist farmers by providing actionable insights, especially for agricultural management and crop health. This ensemble model represents a better solution for advancing agriculture knowledge retrieval by integrating powerful but general LLM with domain-specific knowledge to tackle its unique challenges.