AgriBERT: A Joint Entity Relation Extraction Model Based on Agricultural Text

The core tasks of information extraction are named entity recognition and relationship extraction. These tasks aim to identify and extract entities and their relationships from unstructured text and can be applied to natural language processing tasks, such as knowledge graph construction and intelligent question answering. However, in the agricultural domain, joint extraction tasks of entities and relationships suffer from low accuracy. To address this issue, this paper first constructed a labeled agricultural text entity relationship dataset called AgriRE, which was sourced from authoritative crop pest and disease knowledge websites, with entity and relationship categories designed based on the characteristics of the agricultural texts. The AgriRE dataset includes knowledge related to 28 types of crop diseases and pests, as well as 13 types of relationships associated with these diseases and pests, encompassing a total of 51,458 entity-relation triples. This study then proposed the AgriBERT model, which combined the Robustly Optimized BERT Pretraining Approach (RoBERTa) model and the Conditional Random Field (CRF) model to capture agricultural text features and perform entity-relation extraction. Experiments were conducted using the AgriRE dataset to train and test the model. Experimental results show that the AgriBERT model achieved the highest F1-Score of 97.37% compared to other mainstream models, demonstrating superior performance in joint extraction tasks of entities and relationships in the agricultural domain.