基于主动学习的试油气井控领域命名实体识别模型
    点此下载全文
引用本文:尚福华,马 宁,解红涛.基于主动学习的试油气井控领域命名实体识别模型[J].计算技术与自动化,2022,(2):178-183
摘要点击次数: 57
全文下载次数: 0
作者单位
尚福华,马 宁,解红涛 (东北石油大学 计算机与信息技术学院黑龙江 大庆 163318) 
中文摘要:针对在试油气井控专业领域的命名实体识别任务中,由于没有足够的特征标注数据,使得传统通用领域模型无法高效地进行专业的试油气井控专业领域的命名实体识别的问题,提出了一个基于主动学习方法的试油气井控专业领域命名实体识别模型。该模型首先采用对BERT模型进行的条件预训练,在获取名词向量特性信息后进入双向长短期记忆网络(BiLSTM)中,然后再将输出的特征信息经过条件随机场(CRF)对序列标签的相关性进行约束,最后采用主动学习的方法,筛选出合格的样本进行自动标注后放入已标注数据集中,增加训练样本。实验结果表明在多次迭代训练后,该模型可以在少量标注数据的基础上获得较好的命名实体识别效果并获得较高的命名实体识别准确率。
中文关键词:命名实体识别  深度学习  主动学习  试油气井控领域
 
Well Control Named Entity Recognition Model Based on Active Learning
Abstract:In order to solve the problem that the traditional generic domain model cannot effectively identify named entities in well control for testing oil and gas domain due to insufficient feature label data in the named entity identification task in well control for testing oil and gas domain, a named entity identification model in well control for testing oil and gas domain based on active learning method is presented. The model first adopts the conditional pre training of the Bert model. After obtaining the feature information of the noun vector, it enters the directional Long Short-Term Memory network (Bilstm). Then the output feature information is constrained by the conditional random field (CRF) to constrain the correlation of the sequence labels. Finally, the active learning method is used to screen the qualified samples for automatic labeling and put them into the labeled data set to increase the training samples The experimental results show that the model can achieve better named entity recognition effect and higher named entity recognition accuracy on the basis of a small amount of labeled data after multiple iterative training.
keywords:named entity recognition  deep learning  active learning  well control for testing oil and gas
查看全文   查看/发表评论   下载pdf阅读器