JMIR Publications

ABSTRACT

Background:

Named Entity Recognition (NER) is a key step in clinical natural language processing (NLP). Traditionally, ruled based systems leverage prior knowledge to define rules to identify named entity. Recently, deep learning based NER system become more and more popular and contextualized word embedding, as a new type of representation of the word, has been proposed to dynamically capture word sense using context information and proved successful in many deep-learning based systems in either general domain or medical domain. However, there are very few studies to investigate the effects of combining multiple contextualized embeddings and prior knowledge on the clinical NER task.

Objective:

The aim of the study is to improve the performance of named entity recognizer in the clinical text by combining contextual embedding and prior knowledge.

Methods:

In this study, we investigate the effects of combing multiple contextualized word embedding with classic word embedding in the deep neural networks to predict named entities in the clinical text. We also study if using semantic lexicon could further improve the performance of the clinical NER system.

Results:

As a result, by combining contextualized embeddings such as ELMO and Flair, using our model achieves the F-1 score of 87.30% only using a portion of the 2010 I2b2 NER task dataset. After incorporating the medical lexicon into the word embedding, the F-1 score is further increased to 87.44%. We also found that our model still could achieve the F-1 score of 85.36% when the size of the data is reduced to 40%.

Conclusions:

In conclusion, combined contextualized embedding could be beneficial for the clinical NER task. Moreover, the semantic lexicon could be used to further improve the performance of the clinical NER system. Clinical Trial: NA