A New Method of Named Entity Recognition on Biology Domain
Wande Zhuang, Chengfan Li, Qingjie Zhang, Xuehai Ding, Junjuan Zhao · 2020
It is the phenotypic data of microbial taxonomy that is of great significance. It can be used not only for the identification of microbial taxonomy but also for the study of the interaction between human and the environment. There are bilstm-CRF, Bert-CRF and other models are commonly used for named entity recognition (NER) recently. However, these methods do not consider the sequence information of neurons and do not analyze the hierarchical structure of sentences. In the field of biological phenotypes, there are longer sentences and complex words, the named entity recognition with neural network performs not very well. Based on this, this paper proposed a new NER method which is named BERT-ON-LSTM to identify and extract microbial phenotypic entities from the literature. This paper improved the Bidirectional Encoder Representations from Transformers (Bert) which enabled the model to learn the hierarchical structure information and be more sensitive to words that require long-term memory. And then, conditional random field (CRF) is used to decode since it can be calculated the joint probability distribution of the whole tag sequence. The experimental results on three data sets prove that our model is effective.