KECVS:一个面向专业文献知识实体的类型标注及可视化系统

知识实体的类型标注是专业文献结构化管理和知识脉络挖掘中的一个重要任务。由于专业文献的知识实体具有专业性强、类型多样、随时间变化的特点,如何在无监督的情况下对其进行类型标签抽取、实体类型标注及知识关系挖掘具有重要的意义。设计并实现了一个面向专业文献知识实体的类型标注及可视化系统,提供文献数据的实体识别、实体类型标注、知识实体关系图构建及其可视化等功能,帮助科研工作者更加便捷、直观、准确地把握知识关系和研究热点。     

基于SVM和扩展条件随机场的Web实体活动抽取

在传统信息抽取的基础上,研究Web实体活动抽取,基于格语法对实体活动进行了形式化定义,并提出一种基于SVM(supported vector machine)和扩展条件随机场的Web实体活动抽取方法,能够从Web上准确地抽取实体的活动信息.首先,为了避免人工标注训练数据的繁重工作,提出一种基于启发式规则的训练数据生成算法,将语义角色标注的训练数据集转化为适合Web实体活动抽取的训练数据集,分别训练支持向量机分类器和扩展条件随机场.在抽取过程中,通过分类器获得包含实体活动的语句,然后利用扩展条件随机场对传统条件随机场中不能利用的标签频率特征和关系特征建模,标注自然语句中的待抽取信息,提高标注的准确率.通过多领域的实验,其结果表明,所提出的抽取方法能够较好地适用于Web实体活动抽取.     

融合预训练语言模型和标签依赖知识的关系抽取方法

关系抽取旨在从未经标注的自由文本中抽取实体间的关系.然而,现有的方法大都孤立地预测每一个关系而未考虑关系标签相互之间的丰富语义关联.该文提出了一种融合预训练语言模型和标签依赖知识的关系抽取模型.该模型通过预训练模型BERT编码得到句子和两个目标实体的语义信息,使用图卷积网络建模关系标签之间的依赖图,并结合上述信息指导最...     

基于主动学习的试油气井控领域命名实体识别模型

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。     

用户评论中的标签抽取以及排序

对于一个实体(产品或者商户),往往伴随着成千上万的用户评论。如何从这些冗杂的评论信息中抽取能够描述此实体的精华信息是研究的热点问题。该文提出了一种能够为每个实体抽取特征标签的方法,并且语义去重,保证标签在语义空间内相互独立。首先,对于每个实体的所有评论,进行中文分词、词性标注,并且做依存句法分析。然后,根据每个句子中的依存关系,抽取关键标签,构成此实体的标签库,并且对标签库进行显式语义去重。最后通过K-Means聚类以及Latent Dirichlet Allocation(LDA)主题模型将每个标签映射到语义独立的主题空间,再根据每个标签相对该主题的置信度进行排序。通过以上步骤,可以为每个实体抽取语义独立的关键标签描述,实验中,该文通过对返回标签列表的准确性以及语义多样性进行了统计分析,验证了标签抽取方法的可行性和有效性。     

基于规则推理引擎的实体关系抽取研究

实体关系抽取是指从无结构的自然语言文本中抽取实体之间的语义关系,并以结构化的形式表示出来。传统的实体关系抽取方法只注重一种特定类型的数据源,并需要标注大量的训练数据来训练抽取模型,人工成本高。因此提出了一种综合多种数据源,并结合规则推理引擎的实体关系抽取方法,准确地说就是综合结构化和非结构化两种数据源,在结构化数据提供少量种子的情况下用规则推理引擎推理出更多的实体关系。然后使用远程监督学习方法从无结构的文本中抽取实体关系,通过多次迭代获得最终的实体关系。实验结果证明了该方法的有效性。     

位置标签增强的中文医学命名实体级联识别

针对一般领域的命名实体识别方法不能直接用于中文医学专业实体的识别,现有的相关研究只专注于英文文本和扁平结构的医学实体识别等问题,通过对专业领域实体识别方法的研究,结合中文医学实体的特点提出了一种面向中文医学实体的级联识别方法。将每个字符元素相对于实体的位置标签嵌入模型,并结合中文医学实体跨度内不同元素的重要程度进行实体的融合表示。通过序列标注方法检测字符的位置标签,利用字符的位置信息指导候选实体生成,并进行实体语义分类。模型在CMeEE和CCKS2018数据集以及中文糖尿病科研文献数据集上分别进行扁平实体、嵌套实体和不连续性长实体的识别实验。实验结果表明,该方法能够有效地识别中文医学文本中不同结构的实体。     

多策略中文开放关系抽取方法

开放关系抽取从海量数据中获取知识,是自然语言处理的一个关键技术。开放关系抽取可以实现多种关系的抽取,由于中文领域可供训练的标注数据较少且语义句式较为复杂,面向中文的开放关系抽取存在较多困难。现有的中文开放关系抽取方法存在实体识别覆盖率较低且抽取关系种类单一的问题,无法满足知识图谱扩展等应用需求。该文提出了多策略的开放关系抽取方法,该方法综合利用知识图谱提高了实体识别的覆盖度,依靠实体上下文信息实现了实体对关系的抽取,根据依存句法分析抽取得到全要素三元组,并实现了从文本中抽取实体属性的方法。实验证明,该文的抽取方法准确率高,抽取关系种类多样,可以服务于知识图谱扩展等任务。     

单步片段标注的实体关系联合抽取模型

关系抽取作为知识图谱等诸多领域的上游任务，具有广泛应用价值，近年来受到广泛关注。关系抽取模型普遍存在暴露偏差问题，抽取文本普遍存在实体嵌套和实体重叠问题，这些问题严重影响了模型性能。因此，提出了一种基于片段标注的实体关系联合抽取模型（span-labeling based model,SLM），主要包括：将实体关系抽取问题转化为片段标注问题；使用滑动窗口和三种映射策略将词元（token）序列进行组合排列重新平铺成片段（span）序列；使用LSTM和多头自注意力机制进行片段深层语义特征提取；设计了实体关系标签，使用多层标注方法进行关系标签分类。在英文数据集NYT、WebNLG上进行实验，相对于基线模型F1值显著提高，验证了模型的有效性，能有效解决上述问题。     

面向领域实体关系联合抽取的标注方法

针对传统实体关系标注方法存在效率低下、错误传播、实体冗余等问题,对于某些领域语料中存在“一实体（主实体）同时与多个实体之间存在重叠关系”的特点,提出一种面向领域实体关系联合抽取的新标注方法。首先,将主实体标注为一个固定标签,将文本中与主实体存在关系的其他每个实体标注为对应实体对间的关系类型,这种对实体和关系进行同步标注的方式节省了至少一半的标注成本;然后,直接对三元组进行建模,而不是分别对实体和关系进行建模,通过标签匹配和映射即可获取三元组数据,从而缓解重叠关系抽取、实体冗余以及错误传播等问题;最后,以作物病虫害领域为例进行实验,测试了来自转换器的双向编码器表征量（BERT）-双向长短期记忆网络（BiLSTM）+条件随机场（CRF）端到端模型在1 619条作物病虫害文档的数据集上的性能。实验结果表明该模型的F1值比基于传统标注方式+BERT模型的流水线方法提高了47.83个百分点;与基于新标注方式+BiLSTM+CRF模型、卷积神经网络（CNN）+BiLSTM+CRF等经典模型的联合学习方法相比,该模型的F1值分别提高了9.55个百分点和10.22个百分点,验证了所提标注方法和模型的有效性。     

A SVM based classification method for homogeneous data

A novel classification method based on SVM is proposed for binary classification tasks of homogeneous data in this paper. The proposed method can effectively predict the binary labeling of the sequence of observation samples in the test set by using the following procedure: we first make different assumptions about the class labeling of this sequence, then we utilize SVM to obtain two classification errors respectively for each assumption, and finally the binary labeling is determined by comparing the obtained two classification errors. The proposed method leverages the homogeneity within the same classes and exploits the difference between different classes, and hence can achieve the effective classification for homogeneous data. Experimental results indicate the power of the proposed method.     

Uncertainty sampling-based active learning for protein–protein interaction extraction from biomedical literature

Protein–protein interaction (PPI) extraction from biomedical literature has become a research focus with the rapid growth of the number of biomedical literature. Many methods have been proposed for PPI extraction including natural language processing techniques and machine learning approaches. One problem of applying machine learning approaches to PPI extraction is that large amounts of data are available but the cost of correctly labeling it prohibits its use. To reduce the amount of human labeling effort while maintaining the PPI extraction performance, the paper presents an uncertainty sampling-based method of active learning (USAL) in a lexical feature-based SVM model to tag the most informative unlabeled samples. In addition, some specific samples are ignored to speed up learning process while maintaining desired accuracy. The experiment results on AIMED and CB corpora show that our method can reduce the labeling by 40% and 20%, respectively, without degrading the performance.     

基于聚类分析的不均衡数据标注技术研究

分布不均衡的数据在通过传统聚类分析的方式进行标注时,聚类效果容易偏向于样本数多的类,从而造成标注出现误差的问题。针对此问题提出改进的含有均衡约束聚类算法的标注方法,对不均衡数据的聚类标注准确率实现了比较有效的提高,方法包含数据初始聚类、专家知识调整,数据均衡化处理,含均衡约束聚类等步骤。通过初始聚类对不均衡数据进行初始类标签分配,专家知识调整对部分数据错误标注进行标签调整修改,对数据进行均衡化处理得到均衡数据集,通过均衡约束聚类对均衡数据进行标签最终精确分配。经仿真验证表明,上述方法比较有效的提高了不均衡数据标注准确率。     

面向航空领域HowNet构建的术语内部动态角色标注

术语内部动态角色标注是航空领域HowNet构建的关键环节,其直接影响航空领域HowNet的规模与质量。针对动态角色种类多造成标注困难的问题,提出一种基于KNN的术语内部动态角色标注方法。通过对术语内部词语DEF项的分析进行样本预选择,并在最近邻样本选择阶段融合基于DEF的语义相似性及基于词向量的语境分布相似性。实验结果表明,1-Best、3-Best和7-Best的准确率分别为67.57%、86.00%和94.17%,平均倒数排名MRR为0.7764,优于现有的研究结果。     

A grid density based framework for classifying streaming data in the presence of concept drift

Mining data streams is the process of extracting information from non-stopping, rapidly flowing data records to provide knowledge that is reliable and timely. Streaming data algorithms need to be one pass and operate under strict limitations of memory and response time. In addition, the classification of streaming data requires learning in an environment where the data characteristics might change constantly. Many of the classification algorithms presented in literature assume a 100 % labeling rate, which is impractical and expensive when data records are rapidly flowing in. In this paper, a new incremental grid density based learning framework, the GC3 framework, is proposed to perform classification of streaming data with concept drift and limited labeling. The proposed framework uses grid density clustering to detect changes in the input data space. It maintains an evolving ensemble of classifiers to learn and adapt to the model changes over time. The framework also uses a uniform grid density sampling mechanism to obtain a uniform subset of samples for better classification performance with a lower labeling rate. The entire framework is designed to be one-pass, incremental and work with limited memory to perform any-time classification on demand. Experimental comparison with state of the art concept drift handling systems demonstrate the GC3 frameworks ability to provide high classification performance, using fewer models in the ensemble and with only 4-6 % of the samples labeled. The results show that the GC3 framework is effective and attractive for use in real world data stream classification applications.     

Fast support-based clustering method for large-scale problems

In many support vector-based clustering algorithms, a key computational bottleneck is the cluster labeling time of each data point which restricts the scalability of the method. In this paper, we review a general framework of support vector-based clustering using dynamical system and propose a novel method to speed up labeling time which is log-linear to the size of data. We also give theoretical background of the proposed method. Various large-scale benchmark results are provided to show the effectiveness and efficiency of the proposed method.     

Label noise correction and application in crowdsourcing

The important task of correcting label noise is addressed infrequently in literature. The difficulty of developing a robust label correction algorithm leads to this silence concerning label correction. To break the silence, we propose two algorithms to correct label noise. One utilizes self-training to re-label noise, called Self-Training Correction (STC). Another is a clustering-based method, which groups instances together to infer their ground-truth labels, called Cluster-based Correction (CC). We also adapt an algorithm from previous work, a consensus-based method called Polishing that consults with an ensemble of classifiers to change the values of attributes and labels. We simplify Polishing such that it only alters labels of instances, and call it Polishing Labels (PL). We experimentally compare our novel methods with Polishing Labels by examining their improvements on the label qualities, model qualities, and AUC metrics of binary and multi-class data sets under different noise levels. Our experimental results demonstrate that CC significantly improves label qualities, model qualities, and AUC metrics consistently. We further investigate how these three noise correction algorithms improve the data quality, in terms of label accuracy, in the context of image labeling in crowdsourcing. First, we look at three consensus methods for inferring a ground-truth label from the multiple noisy labels obtained from crowdsourcing, i.e., Majority Voting (MV), Dawid Skene (DS), and KOS. We then apply the three noise correction methods to correct labels inferred by these consensus methods. Our experimental results show that the noise correction methods improve the labeling quality significantly. As an overall result of our experiments, we conclude that CC performs the best. Our research has illustrated the viability of implementing noise correction as another line of defense against labeling error, especially in a crowdsourcing setting. Furthermore, it presents the feasibility of the automation of an otherwise manual process of analyzing a data set, and correcting and cleaning the instances, an expensive and time-consuming task.     

基于置信度的藏文人名识别的主动学习模型研究

训练语料的标注成本是资源稀缺语言处理研究面临的一个重要问题,通过主动学习(active learning)方法可以选择信息量大、无冗余的语料供人工标注,进而大大降低语料标注成本。该文基于CRF模型给出的标注置信度提出了四种主动学习方法,并通过实验确定了这四种主动学习方法的相关参数。实验显示:选择置信度低于0.7的语料进行人工标注,直到新旧模型标注结果的差异度小于0.01%时,仅需6轮迭代;人工标注3.2MB的语料,藏文人名识别的F值可以达到88%,若要达到该识别效果,基于CRF的监督式学习模型需要标注约10MB的语料,该主动学习方法降低了约66%的语料标注规模。     

基于Mask Scoring R-CNN的高质量数据集快速自动标定方法

针对计算机视觉领域人工标定多目标数据集时间冗长的问题,提出一种基于Mask Scoring R-CNN的高质量数据集快速自动标定方法;首先,设计了高质量数据集快速自动标定架构,训练数据自动标定模型并搭建目标分类与标定系统;其次,在对比不同残差网络及引入迁移学习基础上,进一步研究了基于MaskIoU Head的多目标掩膜标定质量评价方法,完成基于Mask Scoring R-CNN的多目标高质量数据集快速自动标定方法设计;最后,以车辆数据为例进行数据集快速自动标定方法验证,实验结果表明,相较于Mask R-CNN和Faster R-CNN方法,Mask Scoring R-CNN方法具有目标数据分类效果好及掩膜分割精度高的优点,检测准确率达到93.4%,且标定速度相较于人工标定速度提升了95.77%。     

基于融合特征相似度的实体消歧方法研究

人名歧义是一种实体身份模糊的现象。在数字资源日益丰富的环境下,人名歧义现象为数据检索带来了很多不确定性,降低了数据检索的准确度。文章运用人名实体的个人信息特征和作者文献话题等特征,采用多特征的融合方法充分挖掘与实体相关联的信息,实现人名消歧。实验使用爬取的某数据库文献数据,进行了人名消歧特征的选取和参数的确定等实验,结果表明提出的方法具有较好的性能。