JMIR Publications

ABSTRACT

Background:

Clinical named entity recognition (CNER), whose goal is to automatically identify clinical entities in electronic medical records (EMR), is an important research direction of clinical text data mining and information extraction. The promotion of CNER can provide support for clinical decision and medical knowledge base construction, and then improve the overall medical quality. Compared with English CNER, due to the complexity of Chinese word segmentation and grammar, Chinese CNER starts late and is more challenging.

Objective:

With the development of distributed representation and deep learning, a series of models have been applied in Chinese CNER. Different from English, Chinese CNER is mainly divided into character-based and word-based methods which cannot make comprehensive use of EMR information and cannot solve the problem of ambiguity in word representation.

Methods:

In this paper, we propose a lattice LSTM model combined with a variant contextualized character representations and CRF layer (ELMo-lattice-LSTM-CRF) for Chinese CNER. Lattice LSTM model can effectively utilize the information of characters and words in Chinese EMR, and the variant ELMo use Chinese characters as input instead of character encoding layer of ELMo model, so as to learn domain-specific contextualized character embeddings.

Results:

Finally, we evaluate our method on two Chinese CNER datasets, one is CCKS-2017 CNER dataset, the other is CCKS-2019 CNER dataset. The results with F1-score are 90.13% and 85.02% on the test sets of these two datasets, respectively.

Conclusions:

It proves that our proposed method is effective in Chinese CNER. Meanwhile, experiments show that variant contextualized character representations can significantly improve the performance of the model.