基于BERT的数控机床故障领域命名实体识别

针对数控机床(computer numerical control,CNC)故障领域命名实体识别方法中存在实体规范不足及有效实体识别模型缺乏等问题,制定了领域内实体标注策略,提出了一种基于双向转换编码器(bidirectional encoder representations from transformers,BERT)的数控机床故障领域命名实体识别方法。采用BERT编码层预训练,将生成向量输入到双向长短期记忆网络(bidirectional long short-term memory,BiLSTM)交互层以提取上下文特征,最终通过条件随机域(conditional random field,CRF)推理层输出预测标签。实验结果表明,BERT-BiLSTM-CRF模型在数控机床故障领域更具优势,与现有模型相比,F1值提升大于1.85%。     

基于双向编码器表示模型和注意力机制的食品安全命名实体识别

针对于目前传统的命名实体识别模型在食品案件纠纷裁判文书领域的准确率不足的问题,在双向长短时记忆网络的基础上提出一种基于双向编码器表示模型(bidirectional encoder representations from transformers,Bert)和注意力机制的命名实体识别模型.模型通过Bert层进行字向量预训练,根据上下文语意生成字向量,字向量序列输入双向长短期记忆网络(bi-directional long short-term memory,BiLSTM)层和Attention层提取语义特征,再通过条件随机场(conditional random field,CRF)层预测并输出字的最优标签序列,最终得到食品案件纠纷裁判文书中的实体.实验表明,该模型在食品纠纷法律文书上面的准确率和F1值分别达到了92.56％和90.25％,准确率相较于目前应用最多的BiLSTM-CRF模型提升了6.76％.Bert-BiL-STM-Attention-CRF模型通过对字向量的预训练,充分结合上下文语意,能够有效克服传统命名实体识别模型丢失字的多义性的问题,提高了食品案件纠纷裁判文书领域命名实体识别的准确率.     

基于多粒度认知的命名实体识别方法

在数据匮乏的领域,命名实体识别效果受限于欠拟合的字词特征表达,引入常规的多任务学习方法可以有所改善,但需要额外的标注成本.针对这一问题,提出了一种基于多粒度认知的命名实体识别方法,在不产生额外标注成本的前提下,增强字特征信息,提高命名实体识别效果.该方法从多粒度认知理论出发,以BiLSTM和CRF为基础模型,将字粒度下的命名实体识别任务与句子全局粒度下的实体数量预测任务相联合,共同优化字嵌入表达.三个不同类型的数据集上的多组实验表明,引入多粒度认知的方法有效地提升了命名实体识别效果.     

面向领域知识图谱的实体关系联合抽取

文本数据中的实体和关系抽取是领域知识图谱构建和更新的来源.针对金融科技领域中文本数据存在重叠关系、训练数据缺乏标注样本等问题,提出一种融合主动学习思想的实体关系联合抽取方法.首先,基于主动学习,以增量的方式筛选出富有信息量的样本作为训练数据;其次,采用面向主实体的标注策略将实体关系联合抽取问题转化为序列标注问题;最后,基于改进的BERT-BiGRU-CRF模型实现领域实体与关系的联合抽取,为知识图谱构建提供支撑技术,有助于金融从业者根据领域知识进行分析、投资、交易等操作,从而降低投资风险.针对金融领域文本数据进行实验测试,实验结果表明,本文所提出的方法有效,验证了该方法后续可用于金融知识图谱的构建.     

针对命名实体识别的数据增强技术

近年来,深度学习方法被广泛地应用于命名实体识别任务中,并取得了良好的效果.但是主流的命名实体识别都是基于序列标注的方法,这类方法依赖于足够的高质量标注语料.然而序列数据的标注成本高昂,导致命名实体识别训练集规模往往较小,这严重地限制了命名实体识别模型的最终性能.为了在不增加人工成本的前提下扩大命名实体识别的训练集规模,本文分别提出了基于EDA(Easy Data Augmentation)、基于远程监督、基于Bootstrap(自展法)的命名实体识别数据增强技术.通过在本文给出的FIND-2019数据集上进行的实验表明,这几种数据增强技术及其它们的组合能够低成本地增加训练集的规模,从而显著地提升命名实体识别模型的性能.     

基于CRF的网页动态关系抽取研究

提出了基于条件随机场(conditional random fields,CRF)的网页动态关系抽取算法.给出了动态关系的定义,建立了动态关系的表示模型,并用一个六维结构来表达动态关系.与传统关系抽取中基于规则或者基于分类的解决方法不同,本文认为可以将动态关系识别问题转化为一个标注问题,并提出了基于CRF的句子层面的关系标注和抽取方法.在本算法中,首先将一个句子通过语义角色标注(semantic role labeling,SRL)系统进行成分识别,然后将语义角色标注结果以及词的POS类型、词组的命名实体类型等作为CRF的训练特征,对句子成分进行标注.最后测试了大量的真实新闻网页,实验结果表明了本文提出算法的实用性和有效性.     

结合原型网络的远程监督命名实体识别方法

针对利用远程监督标注文本实体过程中存在实体类别标注错误问题导致模型难以有效区分各实体的类别特征，影响模型精准度的问题，本文提出一种利用原型网络过滤训练语料中标注错误样本的远程监督命名实体识别方法，利用预训练的原型网络编码正确标注实体生成类别原型表示，过滤语料中距类别原型较远的样本.实验表明，使用原型网络有效地提高了语料的标注质量，提升了模型性能.     

基于BERT和多头注意力的中文命名实体识别方法

针对双向长短时记忆网络-条件随机场(bi-directional long short-term memory-conditional random field,BiLSTM-CRF)模型存在准确率低和向量无法表示上下文的问题，提出一种改进的中文命名实体识别模型。利用裁剪的双向编码器表征模型(bidirectional encoder representations from transformers,BERT)得到包含上下文信息的语义向量;输入双向门控循环单元(bidirectional gated recurrent unit,BiGRU)网络及多头自注意力层捕获序列的全局和局部特征;通过条件随机场(conditional random field,CRF)层进行序列解码标注，提取出命名实体。在人民日报和微软亚洲研究院(Microsoft research Asia,MSRA)数据集上的实验结果表明，改进模型在识别效果和速度方面都有一定提高;对BERT模型内在机理的分析表明，BERT模型主要依赖从低层和中层学习到的短语及语法信息完成命名实体识别(named entity recognition，NER)任务。     

面向武器装备领域的复杂实体识别

针对武器装备领域复杂实体的特点, 提出一种融合多特征后挂载武器装备领域知识的复杂命名实体识别方法。首先, 使用BERT 模型对武器装备领域数据进行预训练, 得到数据向量, 使用Word2Vec模型学习郑码、五笔、拼音和笔画的上下位特征, 获取特征向量。然后, 将数据向量与特征向量融合, 利用Bi-LSTM模型进行编码, 使用CRF解码得到标签序列。最后, 基于武器装备领域知识, 对标签序列进行复杂实体的触发检测, 完成复杂命名实体识别。使用环球军事网数据作为语料进行实验, 分析不同的特征组合、不同神经网络模型下的识别效果, 并提出适用于评价复杂命名实体识别结果的计算方法。实验结果表明, 提出的挂载领域知识且融合多特征的武器装备复杂命名实体识别方法的F1值达到95.37%, 优于现有方法。     

多神经网络协作的军事领域命名实体识别

互联网公开数据蕴含着大量高价值的军事情报,成为获取开源军事情报的重要数据源之一。军事领域命名实体识别是进行军事领域信息提取、问答系统、知识图谱等工作的基础性关键任务。相比较于其他领域的命名实体,军事领域命名实体边界模糊,界定困难;互联网媒体中军事术语表达不规范,随意性的简化表达现象较普遍;现阶段面向军事领域的公开语料鲜见。该文提出一种考虑实体模糊边界的标注策略,结合领域专家知识,构建了基于微博数据的军事语料集MilitaryCorpus;提出一种多神经网络协作的军事领域命名实体识别模型,该模型通过基于Transformer的双向编码器(bidirectional encoder representations from transformers, BERT)的字向量表达层获得字级别的特征,通过双向长短时记忆神经网络(bi-directional long short-term memory, BiLSTM)层抽取上下文特征形成特征矩阵,最后由条件随机场层(conditional random field, CRF)生成最优标签序列。实验结果表明:相较于基于CRF的实体识别模型,应用该文提出的BERT-BiLSTM-CRF模型召回率提高28.48%,F值提高18.65%;相较于基于BiLSTM-CRF的实体识别模型,该文模型召回率提高13.91%,F值提高8.69%;相较于基于CNN (convolutional neural networks)-BiLSTM-CRF的实体识别模型,该文模型召回率提高7.08%,F值提高5.15%。     

Features of Women's language

Language is a means of verbal communication. People use language to communicate with each other. In the society, no two speakers are exactly alike in the way of speaking. Some differences are due to age, gender, statue and personality. Above all, gender is one of the obvious reasons. The writer of this paper tries to describe the features of women＇s language from these perspectives： pronunciation, intonation, diction, subjects, grammar and discourse. From the discussion of the features of women＇s language, more attention should be paid to language use in social context. What＇s more, the linguistic phenomena in a speaking community can be understood more thoroughly.     

Jane Eyre——A Feminist In The 19th Century

In the 19th century the society was controlled by men, and women were just appendants of them, they had not any rights and freedom. But Jane was an exception, she showed some characteristics of early feminist. Jane showed her characteristics of feminism in three aspects: rebellion, equality, and independence. These characteristics were helpful to her success, and feminism is the only way out for women of that time.     

Editorial note for the special topic on ＂Nanopore Analysis＂

The concept of nanopore analysis, using the pore-forming protein a-hemolysin to detect individual nucleic acids at a single-molecule level, was first proposed in 1996. Over the past two decades, tremendous progress has been made in the nanopore field, and nanopore analysis has become a label-free and high-throughput method for probing bio- molecules and other analytes with single-molecule sensi- tivity, especially holds the promising for ＂third generation＂ DNA sequencing. However, challenges still remain in the experimental strategies and the design of whole nanopore-based instruments. Here, we proudly present a special topic dedicated to the topic of ＂Nanopore Analysis＂, with 8 reviews/articles providing up to date coverage of the experimental strategies, theoretical calcu- lations and simulations, and instrument design. Reviews and articles on the experimental strategies cover control of DNA partitioning into a nanopore, detection of target DNA, and the advantages of nanopore-based DNA sequencing. The theoretical calculations and simulations discuss the translocation behavior of DNA, and an inte- grated measurement system and data analysis software are presented for instrument design.     

Effects of climate change on biodiversity

1 Rise of studies on climate change＇s effects on biodiversity
Until the 1980s, climate change and biodiversity were studied as two independent disciplines for more than a century. In 1992, the Ecological Society of America＇s annual report named climate change, biodiversity, and the sustainable ecological system as the three major global environmental issues of the twenty-first century [1].     

Discussing the Expansive Learning in Modern Organizations

<正>1 Introduction Endeavours on learning methods have long occupied the mind spaces of scholars and there have been theories abound on how people learn and in what way organisations can use some of the knowledge gained in the art of learning to maximise human productivity.While the theories by themselves are not solutions to problems,they are steps of inquiry that provide us with interesting perspectives that we can use to extend our knowledge of the dimensions of learning.This paper is an exploratory work on the theories postulated by three notable scholars including Zimmerman     

Progress in Linear Scaling Non-Orthogonal Localized Molecular Orbitals Method

The non-orthogonal localized molecular orbital(NOLMO)is the most localized representation of electronic degrees of freedom.As such,NOLMOs are thus potentially the most efficient for linear scaling calculations of electronic structures for large systems.However,direct ab initio calculations with NOLMO have not been fully implemented and widely used,partly because of the slow convergence issue in the optimization of NOLMO.We devel-     

An Accurate and Universal HEOM Approach to Strongly Correlated Quantum Impurity Systems

A hierarchical equations of motion(HEOM)approach is developed for general open quantum systems coupled to fermionic environment.The HEOM method is in principle formally exact,as it resolves nonperturbatively the combined effects of many-body interaction,system-bath dissipation,and non-Markovian memory.In practice,the HEOM approach is highly accurate and efficient for the characterization of strongly correlated quantum impurity sys-     

关于有理混合吸引子的Ciric型非唯一不动点定理

本文我们应用有理混合吸引子条件去证明具非唯一不动点的映射Ciric型的若干不动点定理.结果推广并改进已知的一些结果.     

次线性算子交换子在变指数Herz-Morrey空间上的有界性

本文主要利用给出的次线性算子分别与BMO函数及Lipschitz函数生成的交换子在变指数L~(p(·))(R~n)空间上的有界性,证明了其在变指数Herz-Morrey空间MK_(q,p(·))~α~((·)),λ(R~n)上的有界性.     

利用GPS双频观测值实现PPP电离层延迟高阶项改正

给出了利用双频观测值计算L3组合电离层延迟高阶项改正的方法,并与全球电离层延迟文件的改正效果进行对比.利用赤道附近的15个国际全球卫星导航定位系统服务组织(IGS)站的数据进行比较,结果表明:2种方法计算的电离层二阶项延迟互差最大不超过1 cm,三阶项延迟互差最大不超过5 mm;电离层高阶项改正后的观测值精密单点定位(PPP)解算结果N、E、U方向互差平均值分别为0.4、0.5、1.0mm,因此2种改正方法效果在同一水平.