BiLSTM-LAN based Medical Named Entity Recognition
Chunling Chen, Qianli Niu, Yongan Guo · 2020
With the wide application of information technique in the medical field, various medical information systems have produced massive medical data. The problem of mining knowledge from medical data has been extensively studied. For the medical named entity recognition problem, Bidirectional Long-Short-Term Memory model with Conditional Random Field (biLSTM-CRF) is commonly used. However, CRF cannot capture long term output label relations because of the Markov assumptions. To solve this problem, we propose to use a bidirectional Long-Short-Term Memory model with Label Attention Network (biLSTM-LAN) for medical named entity recognition. The experimental results show that using LAN to replace CRF can improve the recognition accuracy when their number of parameters are similar.