Advanced Agricultural Query Resolution Using Ensemble-Based Large Language Models

Effective knowledge retrieval is crucial for addressing challenges related to optimization, such as pest management, soil health and crop productivity. Current single-model approaches struggle with limited accuracy, inconsistent responses, and inability to handle the increasing complexity of agricultural data, leading to unreliable recommendations for farmers. This study demonstrates an innovative weighted voting ensemble method that improves agricultural knowledge retrieval by combining Meta-LLaMA 3.1, Agricultural-BERT, and BERT-based-uncased. The ensemble model optimizes the prediction process by utilizing domain-specific data and a weighted voting mechanism to improve query performance and answer production. Our study outperformed individual models in providing accurate and contextually relevant responses, with an accuracy of 93%. We evaluated the system using both BLEU and ROUGE metrics to assess the quality of the generated text. Our ensemble model achieved a BLEU score 53.8 and demonstrated superior performance in the ROUGE evaluation, with ROUGE-1, ROUGE-2, and ROUGE-L scores of 0.70, 0.55 and 0.61. These results highlight the method's ability to generate contextually suitable, domain-specific responses that address the practical needs of agricultural experts. By integrating advanced LLMs with domain-specific knowledge, the proposed ensemble system significantly improves agricultural knowledge retrieval and provides more accurate and practical responses in the field. The findings suggest that the ensemble approach can effectively support decision-making in agricultural practices, particularly in management agricultural optimization.