Abstract

Named entity recognition (NER) is one of the foundational and key tasks of knowledge graph construction. Traditional entity recognition methods of diseases and insect pests highly rely on artificial design features. To solve this problem, this paper constructs a network model combining bidirectional long short-term memory network (BiLSTM) and conditional random field (CRF). In addition, BIOES labeling strategy is used to solve the problem that the boundary of diseases and pests’ entities is not clear. First, the text data of grape diseases and insect pests were annotated in sequence with BIOES method. Secondly, the labeled text sequence was vectorized at character level by word vector technology. The BiLSTM network was used to extract the context feature information of the text sequence. Finally, the CRF model was used to constrain the label rules at the sentence level to identify entities. The experimental results show that the BiLSTM-CRF model has a good effect on grape disease entity recognition, and the F1 value is as high as 95.44%. Compared with other comparison models, the recognition effect of the BiLSTM-CRF model is relatively stable, which to a certain extent verifies that the model has better recognition performance in the task of identifying grape diseases and pests’ entities.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call