中文电子病历命名实体识别的主动学习方法研究

摘要

主动学习通过选择现有模型中未充分训练的数据进行迭代训练,从而利用少量标注数据训练出较高性能的模型.传统的主动学习方法只关注数据本身的信息量而忽略了该数据是否孤立.论文提出一种改进的主动学习方法,利用相似病历中出现的词汇往往雷同的特点,以文档中文字的分布衡量其在样本集的普遍程度,并以此对信息量进行加权.而后实现该方法,并进行了与消极学习和传统主动学习方法的对比实验.结果表明,该方法相对消极学习与传统的主动学习方法,效果有明显提升,能够减少对标注数据的需求.%Active learning can be trained iteratively by selecting incompletely trained data from the existing model,thus training model with a higher performance by using a small amount of labeled data.Traditional active learning methods focus only on the amount of information in the data itself,and ignore whether the data is isolated.This paper proposes an improved active learning method,uses the similar vocabulary in similar medical records,measures its prevalence in the sample set with the distribution of text in the document,and weights the amount of information.And then the method is implemented,and the comparative experiment of passive learning and traditional active learning method is done.The results show that the method significantly improves the effect of passive learning and traditional active learning method,and it can reduce the demand for labeled data.