GCN with External Knowledge for Clinical Event Detection

GCN with External Knowledge for Clinical Event Detection Dan Liu, Zhichang Zhang†, Hui Peng, Ruirui Han College of Computer Science and Engineering, Northwest Normal University, China 3394055228@qq.com, zzc@nwnu.edu.cn penghuipwld@163.com, 1090400220@qq.com Abstract In recent years, with the development of deep learning and the increasing demand for medical information acquisition in medical information technology applications such as clinical decision support, Clinical Event Detection has been widely studied as its subtask. However, directly applying advances in deep learning to Clinical Event Detection tasks often produces undesirable results. This paper proposes a multi-granularity information fusion encoder-decoder framework that introduces external knowledge. First, the word embedding generated by the pre-trained biomedical language representation model (BioBERT) and the character embedding generated by the Convolutional Neural Network are spliced. And then perform Part-of-Speech attention coding for character-level embedding, perform semantic Graph Convolutional Network coding for the spliced character-word embedding. Finally, the information of these three parts is fused as Conditional Random Field input to generate the sequence label of the word. The experimental results on the 2012 i2b2 data set show that the model in this paper is superior to other existing models. In addition, the model in this paper alleviates the problem that “occurrence” event type seem more difficult to detect than other event types. 1 Introduction Electronic medical records are an inevitable product of medical information. The use of natural language processing (NLP) technology to effectively detect clinical events in electronic medical records becomes crucial as the number of electronic medical records rapidly grows. It has been widely studied because of its potential help in constructing clinical event lines, medical Q & A, assisted diagnosis and other tasks. The task of Clinical Event Detection (CED) is to identify the boundary of the event in the electronic medical record and determine its type. The event detection to identify the boundary and determine type is usually considered as a sequence labeling task. The emergence of deep learning has greatly improved the performance of the sequence labeling task model. Bidirectional long short-term memory network (BiLSTM) is widely employed in sequence la- beling tasks owing to its high power to learn the contextual representation of words. Huang et al. (2015) was the first to apply the bidirectional long short-term memory (BiLSTM) and the conditional random field (CRF) to sequence labeling tasks. But BiLSTM needs to be processed sequentially over time, it cannot be calculated in parallel. Instead, the Transformer not only advantage in modeling the long-range context, but also fully make use of the concurrence power of GPUs. Yan et al. (2019) found that due to its position coding problem, the performance of Transformer in sequence labeling tasks is not as good as in other NLP tasks, and solves the position coding problem. However, due to the lack and particularity of clinical data, the performance of directly applying these technological advances to CED tasks is not ideal. For this reason, we need to use a large amount of medical information for tokens representation. In the past, Word2Vec (Mikolov et …

Content truncated to 500 words