Citrus Knowledge Extraction Model Based on Qwen2.5

In response to the need for improved model adaptability and recognition accuracy in knowledge extraction for the citrus domain, this study proposes a knowledge extraction method based on the Qwen2.5-3B large language model and the combined fine-tuning approach of Prompt Learning and LoRA(Low-Rank Adaptation). By incorporating the Qwen2.5-3B model, which contains 13 billion parameters, along with the semantic guidance of Prompt Learning and the low-rank adaptive fine-tuning strategy of LoRA, the method effectively enhances the model's accuracy in identifying professional citrus entities and agricultural entity relationships. The experiments utilized a professional dataset constructed from citrus pest and disease maps and cultivation manuals. The results demonstrate that this method significantly outperforms traditional fine-tuning techniques in both named entity recognition and relation extraction tasks, validating the effectiveness and generalization capability of prompt learning and parameter-efficient fine-tuning techniques in the field of agricultural knowledge extraction.