Integrating fine-tuning and retrieval-augmented generation to address the application challenges of pollution control...

Environmental pollution has become increasingly complex, prompting governmental agencies to issue numerous technical and detailed guidelines for pollution control. While these documents contain substantial domain-specific knowledge, their complexity often impedes effective implementation and broader public comprehension. This study presents the first systematic application of combining Retrieval-Augmented Generation (RAG) with fine-tuning (FT) in the environmental domain, with the goal of enhancing the interpretation and generation of pollution control texts. The proposed hybrid system was evaluated on documents from local environmental agencies, using a combination of automated metrics (e.g., ROUGE, BLEU, and BERTScore) and preliminary expert assessment. The results indicate that the hybrid model achieved higher scores than other models on both automated metrics and expert evaluations. This suggests that it can effectively interpret and coherently generate technical guidance related to pollution control. These findings demonstrate the model's potential to support more reliable policy comprehension and implementation in environmental management, contributing to the intelligent interpretation of environmental policies and the advancement of data-driven environmental governance.