基于神经网络的医疗命名实体抽取研究

Research on Medical Named Entity Extraction Based on Neural Network

随着互联网技术的快速发展和信息化时代的到来,人们对数据的需求越来越大。如何从海量文本中快速、准确地获取有用信息成为了一个亟待解决的问题。作为一种重要的自然语言处理方法,命名实体识别(NER)在很多领域都有广泛应用,如机器翻译、自动问答系统等。但是由于医学文献本身专业性较强,且存在大量的半结构化或非结构化文本,使得传统的NER模型很难直接用于医学领域。因此,文章针对医学文本特点,提出了一种基于BiLSTM-CRF的医疗命名实体抽取算法,并通过实验证明了该方法能够有效提高医疗命名实体识别率。重点分析了BiLSTM-CRF模型原理及其优势所在;然后,将BiLSTM-CRF模型与改进后的BiLSTM模型相结合,构建出一种新的双向长短期记忆网络模型——BiLSTM-CRF+BiLSTM模型;最后,利用上述模型进行训练,实现了医疗命名实体抽取。

With the rapid development of Internet technology and the advent of the information age,people's demand for data is growing.How to obtain useful information quickly and accurately from massive texts has become an urgent problem to be solved.As an important natural language processing method,named entity recognition(NER)has been widely used in many fields,such as machine translation,automatic question answering system,etc.However,due to the strong professional nature of medical literature and the existence of a large number of semi-structured or unstructured texts,it is difficult to directly apply the traditional NER model to the medical field.Therefore,according to the characteristics of medical texts,this paper proposes a medical named entity extraction algorithm based on BiLSTM-CRF,and proves that this method can effectively improve the recognition rate of medical named entities through experiments.Firstly,the background and significance of the subject are introduced,and the relevant research status at home and abroad is summarized.Secondly,the basic concept of named entity recognition and common feature selection methods are described in detail.Thirdly,it focuses on the theory and advantages of BiLSTM-CRF model.Then,by combining Bilstm-CrF model with the improved BiLSTM model,a new bidirectional long and short term memory network model,Bilstm-Crf+BiLSTM model,is constructed.Finally,medical named entity extraction is realized by using the above model for training.