基于BLSTM的临床文本实体关系抽取

Clinical Text Entity Relationship Extraction Based on BLSTM

实体关系的提取是构建知识库的重要组成部分,对临床文本实体关系的研究可以促进医疗卫生的发展。传统针对实体关系抽取的方法大多是基于规则或是机器学习,需要领域专家来制定大量特征,而且特征的多少和准确性同时影响关系抽取结果的准确性。为了能更好的提取文本特征,同时减少手工制造特征带来的麻烦,该文提出使用双向长短期记忆网络(BLSTM),利用该模型提取句子级语义特征,从而达到更好的实体关系抽取效果。通过对比其他模型,证实了该模型的有效性。

The extraction of entity relationships is an important part of building a knowledge base.The study of the relationship between clinical text can promote the development of health care.Traditional methods for entity relationship extraction are mostly based on rules or machine learning.Domain experts are required to manufacture a large number of features,the number and accuracy of features affect the accuracy of the relationship extraction results. In order to extract more effective text features and reduce the trouble caused by hand-made features,this paper proposes to use bidirectional long short-term memory network (BLSTM) to extract sentence-level semantic features, so as to achieve better entity relationship extraction.The validity of the model was confirmed by comparing other models.