|
|
FCS | 文章解读:弱监督槽位填充中极大噪声样本的自动诊断 |
|
论文标题:Combating with extremely noisy samples in weakly supervised slot filling for automatic diagnosis(弱监督槽位填充中极大噪声样本的自动诊断)
期刊:Frontiers of Computer Science
作者:Xiaoming SHI, Wanxiang CHE
发表时间:15 Oct 2023
DOI:10.1007/s11704-022-2134-1
微信链接:点击此处阅读微信文章
原文信息
标 题:
Combating with extremely noisy samples in weakly supervised slot filling for automatic diagnosis
发表年份:
2023年
原文链接:
https://journal.hep.com.cn/fcs/EN/10.1007/s11704-022-2134-1
引用格式:
Xiaoming SHI, Wanxiang CHE. Combating with extremely noisy samples in weakly supervised slot filling for automatic diagnosis. Front. Comput. Sci., 2023, 17(5): 175333
01
导读
槽填充是对话系统自动诊断的一个至关重要的模块,可以为特定类型的信息(槽)提取实体。医生的回答可以看作是对患者询问的弱监督,通过这种方式,可以从未标记的诊断对话中获得大量弱标记数据,缓解数据标注成本高、耗时长的问题。然而,弱标记数据受到样本噪声极大的影响,为了解决这一问题,本文提出了一种简单有效的协同弱教学方法。该方法同时训练两个槽填充模型,这两个模型从两个不同的弱标记数据中学习,然后,一个模型迭代地利用另一个模型生成的选定的弱标记数据。该模型由弱标注数据上的Co-Weak-Teaching得到,可以直接在测试数据上进行测试,也可以在少量人工标注数据上进行顺序微调。在这两种设置下的实验结果表明了该方法的有效性,微观和宏观分数分别提高了8.03%和14.74%。
02
方法介绍
术语定义
定义1.专家级标注数据集
专家级标注数据集表示为
,其中
表示专家级标注数据集中从在线医疗社区收集的第i个样本中的用户查询,
表示专家给出的专家级标注标签
,
表示专家级标注数据集的数量。
定义2.未标记的数据集
将未标记数据集记为
,其中
表示从在线医疗社区收集的未标记数据集中的第
个样本中的用户查询,
表示对应的医生对
的响应,
表示未标记数据集的数量。
定义3.弱标签
表示
的弱标签,通过与MKB中的医学概念进行精确字符串匹配,从
中被提取。
定义4.弱标签数据集
弱标签数据集表示为
,其中
为未标记数据集中的用户查询,
为对应的弱标签。
Co-Weak-Teaching
本文提出了一种简单而有效的方法,称为“Co-Weak-Teaching”,它允许用极具噪声的标签鲁棒地训练深度网络。本文的方法同时维护两个网络,这两个网络使用两个不同的弱标记数据初始化,以确保模型从两个不同的方面学习任务,总体架构如图1所示。
![](data:image/png;base64,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)
图1 模型框架
本文提出的协同弱学习方法用于从极具噪声的弱标记数据中学习,弱标记数据被分为两个独立的数据,弱标记数据1和弱标记数据2。模型1和模型2用预训练的BERT模型初始化,这两个模型交换选定的弱标记数据用于模型训练。
模型初始化
BERT在各种任务上取得了最先进的表现,因此,在这项工作中,两个深度网络都使用公开可用的预训练BERT进行初始化,这两个模型分别表示为图1中的模型1和模型2。
训练
为了保证这两个模型之间的差异性,这两个模型被馈送不同的弱标记数据。具体而言,将整个弱标记数据随机分为两个没有共同样本的数据集
和
,
(算法1中的第一行)。模型分别在这两个数据集上进行训练,并独立更新。
算法的输入是两个分类器,记为
和
,其中
和
分别为这两个函数的权值,
为输入的查询令牌,
为令牌数量,
为令牌嵌入维数。
模型训练被进行。首先,
被随机分为两个子数据
和
。其次,
和
分别对
和
进行训练(算法1中的第4行和第5行)。形式上,对于每个在
中的样本
,
![](data:image/png;base64,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)
式中
为
中的样本个数,
表示第i个输入患者查询,
表示第
个对应的弱标签,
表示交叉熵损失。
对于每个在
中的样本
,
![](data:image/png;base64,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)
式中
为
中的样本个数,
表示第i个输入患者查询,
表示第i个对应的弱标签,
表示交叉熵损失。
亚弱标记数据采集
为了获得亚弱标记数据,训练后的模型使用训练后的
和
(算法1中的第6行和第7行)从弱标记数据
中选择具有高置信度的样本。损失用作置信度度量,选择顶部损耗小的样品。形式上,对于所有弱标记数据中的每个样本
,
![](data:image/png;base64,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)
式中
为随迭代次数
增长的选择率,
为损失函数,
为亚弱标记数据1。请注意,这里
用于共同训练。
对亚弱标记数据2通过用
同样的方法计算,
![](data:image/png;base64,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)
式中
为亚弱标记数据2。
联合教学
联合教学旨在同时培养两种模式,从两个方面学习任务,这两个模型相互交换选定的数据。需要注意的是,这两个数据是以协同教学的方式交替替换的,如算法1中的第8行和第9行所示。为了更好地从弱标记数据中学习,训练过程和亚弱标记数据采集被执行了
次。
每次回合后(算法1中的第2行和第3行)更新选择率
,
![](data:image/png;base64,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)
其中
为迭代次数,
是一个固定值,也是一个预定义的固定值。随着训练次数的增加,该方法将保留越来越多的弱标记数据。最后,放弃弱标记数据的
,将其视为极端数据。该方法采用先学习简单样本后学习难样本的课程学习方式进行学习,课程学习方式有助于模型更好地学习数据,选择在弱监督数据中表现较好的模型进行测试。
算法1:Co-Weak-Teaching
![](data:image/png;base64,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)
03
主要贡献
1.本文提出了一种简单有效的方法,称为Co-Weak-Teaching,以对抗极端噪声的样本。
2.本文进行了大量的实验,证明了联合教学的有效性。
3.本文实验结果证明了仅从医生的反应中学习是有研究空间的。
04
实验结果的简单总结
表1列出了实验的主要结果,报告了在专家标记数据集上进行微调和不进行微调时,在设置下的精度、召回率、微观
、宏观
和转向精度结果(%)。弱标记数据有三种使用方法:(1)Naive (2)co-teaching (3) Co-Weak-Teaching(本文所提出的)。
通过对比Co-Weak-Teaching预训练和朴素预训练的结果可以看出,在不进行微调设置的情况下,Co-Weak-Teaching可以使模型在微观精度和宏观精度上分别提高22.03%和19.03%。在微调设置下,Co-Weak-Teaching也达到了最佳性能,微精度提高了0.92%,宏观精度提高了1.22%。结果表明,Co-Weak-Teaching能够显著提高学生在任务上的表现。与不进行微调的情况相比,微调方法在微观精度和宏观精度上都有提高,表明了微调过程的有效性。
此外,模型在微观精度上的表现要优于宏观精度,原因是数据集的长尾效应严重。统计表明,数据集中存在许多少射甚至零射标签。结果表明,伪数据集为低频标签提供了更多的样本,显著提高了宏精度。
![](/upload/paper/images/2024/2/2024211540315550.png)
表1 在带注释的数据集上进行微调和不进行微调的精度、召回率、微观、宏观和转向精度结果(%)
05
与其他相关研究的对比
DS-AUDI在简化诊断程序和降低收集患者信息的成本方面有巨大的潜力和诱人的技术价值,而DS-AUDI存在一个基本障碍,即患者生成查询的插槽填充(SF)。关于如何从极具噪声的样本中学习SF模型这个难题,本文提出的方法相较于其他研究更加简单有效,这种方法即使在极度嘈杂的样本中也能鲁棒地训练深度网络。
解读:戴西件 南昌大学第二附属医院
审核:张 琨 合肥工业大学
![](/upload/paper/images/2024/2/202421154156800.png)
Frontiers of Computer Science
Frontiers of Computer Science (FCS)是由教育部主管、高等教育出版社和北京航空航天大学共同主办、SpringerNature 公司海外发行的英文学术期刊。本刊于 2007 年创刊,双月刊,全球发行。主要刊登计算机科学领域具有创新性的综述论文、研究论文等。本刊主编为周志华教授,共同主编为熊璋教授。编委会及青年 AE 团队由国内外知名学者及优秀青年学者组成。本刊被 SCI、Ei、DBLP、INSPEC、SCOPUS 和中国科学引文数据库(CSCD)核心库等收录,为 CCF 推荐期刊;两次入选“中国科技期刊国际影响力提升计划”;入选“第4届中国国际化精品科技期刊”;入选“中国科技期刊卓越行动计划项目”。
《前沿》系列英文学术期刊
由教育部主管、高等教育出版社主办的《前沿》(Frontiers)系列英文学术期刊,于2006年正式创刊,以网络版和印刷版向全球发行。系列期刊包括基础科学、生命科学、工程技术和人文社会科学四个主题,是我国覆盖学科最广泛的英文学术期刊群,其中12种被SCI收录,其他也被A&HCI、Ei、MEDLINE或相应学科国际权威检索系统收录,具有一定的国际学术影响力。系列期刊采用在线优先出版方式,保证文章以最快速度发表。
中国学术前沿期刊网
http://journal.hep.com.cn
![](/upload/paper/images/2024/2/2024211541379470.png)
特别声明:本文转载仅仅是出于传播信息的需要,并不意味着代表本网站观点或证实其内容的真实性;如其他媒体、网站或个人从本网站转载使用,须保留本网站注明的“来源”,并自负版权等法律责任;作者如果不希望被转载或者联系转载稿费等事宜,请与我们接洽。