Naming entity recognition of citrus pests and diseases based on the BERT-BiLSTM-CRF model

In the agricultural industry, there is still a need for improved systematic integration of knowledge related to citrus pests and diseases. Creating a knowledge base focused on citrus pests and diseases and developing a knowledge map would enable the formation of a system providing accurate and convenient prevention and control methods for citrus growers. However, the majority of existing data on citrus pests and diseases is unstructured, making identifying named entities from large volumes of unstructured text data particularly important for creating a knowledge map. In this paper, we propose a training model based on Bidirectional Encoder Representation from Transformers (BERT), combined with Bidirectional Long and Short Term Memory Networks (BiLSTM) and Conditional Random Field (CRF), to extract specific entity categories from unstructured data. When tested on a created dataset, the model achieved an accuracy rate of 0.9423 and an f1 value of 0.8048. The results demonstrate that the proposed method can be applied to extract specific entity categories from text data related to citrus pests and diseases, laying a solid foundation for the subsequent construction of a knowledge map. This paper introduces a method for named entity recognition in the field of citrus pests and diseases, which could serve as a reference for constructing knowledge maps in other fields and improving the utilization of domain knowledge.