【摘要】 文本命名实体识别是信息抽取和预测的基本与关键任务,提出基于深层网络模型的命名实体识别方法,构建多种学习模型.首先对文本进行清洗并规范化,生成基本结构和表示方法,结合边界特征构建深层条件随机场模型,选择最优特征集训练.将文本表示为词向量形式,以向量作为深层神经网络的输入进行模型的训练,提出了基于块表示的BR-BiRNN、BR-BiLSTM-CRF命名实体识别深层网络模型,在I2B2 2006年和2014年评测数据集及妇产科真实医疗文本上实验,结果均比传统的SVM、HM M、CRF的F值高.更多还原

【Abstract】 Text named entity recognition is the basic and key task of information extraction and prediction. The named entity recognition method based on deep network model is proposed,and then we build several learning models. First,the text is cleaned and normalized,basic structure and representation methods are generated,and a deep conditional random field model is built with boundary features,then we choose the optimal feature set to train. The text is represented as a word vector form,and the vector is used as the input of the deep neural network to train the model. We propose the BR-BiRNN,BR-BiLSTM-CRF deep network model for named entity recognition based on block representation,do experiment on the I2B2 2006 and 2014 evaluation datasets and gynecological real medical text,the results are higher than the traditional SVM,HMMand CRF on F value.更多还原