检索结果-南通市图书馆

维普期刊数据库博看期刊

融合相似度负采样的远程监督命名实体识别方法

在线全文

学校读者我要写书评

暂无评论

计算机应用研究 2024年第8期41卷 2322-2328页

作者：刘杨线岩团相艳黄于欣昆明理工大学信息工程与自动化学院昆明650500 云南省人工智能重点实验室昆明650500

实体漏标是目前远程监督命名实体识别(distantly supervised named entity recognition,DS-NER)存在的一个难点问题。训练集中的漏标实体在模型训练中提供了不正确的监督信息,模型将在后续预测实体类型时更倾向于将该类实体预测为非实体... 详细信息

实体漏标是目前远程监督命名实体识别(distantly supervised named entity recognition,DS-NER)存在的一个难点问题。训练集中的漏标实体在模型训练中提供了不正确的监督信息,模型将在后续预测实体类型时更倾向于将该类实体预测为非实体,导致模型的实体识别和分类能力下降,同时影响了模型的泛化性能。针对这一问题,提出了融合实体特征相似度计算负采样命名实体识别方法。首先,通过对候选样本和标注实体样本进行相似度计算并打分;其次,以相似度得分作为依据对候选样本进行采样,采样出参与训练的样本。与随机负采样方法相比,该方法通过结合相似度计算,降低了采样到漏标实体的可能性,进而提高了训练数据的质量,从而提升了模型的性能。实验结果表明,该方法在CoNLL03、Wiki、Twitter三个数据集上与其他模型相比,比基线模型平均取得了5%左右的F_(1)值提升,证明了该方法能够有效缓解远程监督条件下实体漏标带来的命名实体识别模型性能下降的问题。

关键词：命名实体识别实体漏标远程监督负采样数据增强

Text Detection in Natural Scene Images Using Morphological Component Analysis and Laplacian Dictionary

在线全文

学校读者我要写书评

暂无评论

IEEE/CAA Journal of Automatica Sinica 2020年第1期7卷 214-222页

作者： Shuping Liu yantuan xian Huafeng Li Zhengtao Yu School of Information Engineering and Automation and Key Laboratory of Intelligent Information Processing Kunming University of Science and Technology

Text in natural scene images usually carries abundant semantic information. However, due to variations of text and complexity of background, detecting text in scene images becomes a critical and challenging task. In this paper, we present a novel method to detect text from scene images. Firstly, we decompose scene images into background and text components using morphological component analysis(MCA), which will reduce the adverse effects of complex backgrounds on the detection *** order to improve the performance of image decomposition,two discriminative dictionaries of background and text are learned from the training samples. Moreover, Laplacian sparse regularization is introduced into our proposed dictionary learning method which improves discrimination of dictionary. Based on the text dictionary and the sparse-representation coefficients of text, we can construct the text component. After that, the text in the query image can be detected by applying certain heuristic rules. The results of experiments show the effectiveness of the proposed method.

关键词： Dictionary learning Laplacian sparse regularization morphological component analysis(MCA) sparse representation text detection

Linguistic feature template integration for Chinese-Vietnamese neural machine translation

在线全文

学校读者我要写书评

暂无评论

Frontiers of Computer Science 2022年第3期16卷 223-225页

作者： Zhiqiang YU yantuan xian Zhengtao YU Yuxin HUANG Junjun GUO Faculty of Information Engineering and Automation Kunming University of Science and TechnologyKunming 650500China School of Mathematics and Computer Science Yunnan Minzu UniversityKunming 650500China Yunnan Key Laboratory of Artificial Intelligence Kunming University of Science and TechnologyKunming 650500China

1 Introduction and main contributions Template-based approaches have achieved significant progress in low-resource neural machine translation(NMT)recently[1],such as the efficient works,NMT-GTM[2],SoftPrototype[3],***,most previous works only retrieve target sentence as template to generate translation,neglecting the utilization of linguistic feature that contained in the source sentence and template.

关键词： translation template such

在线全文

学校读者我要写书评

暂无评论

基于伪触发词的并行预测篇章级事件抽取方法

电子技术应用 2024年第4期50卷 67-74页

作者：秦海涛线岩团相艳黄于欣昆明理工大学信息工程与自动化学院云南昆明650500 昆明理工大学云南省人工智能重点实验室云南昆明650500

篇章级事件抽取一般将事件抽取任务分为候选实体识别、事件检测和论元识别3个子任务,然后采用级联的方式依次进行,这样的方式会造成误差传递;另外,现有的大多数模型在解码事件时,对事件数量的预测隐含在解码过程中,且只能按照预定义的... 详细信息

篇章级事件抽取一般将事件抽取任务分为候选实体识别、事件检测和论元识别3个子任务,然后采用级联的方式依次进行,这样的方式会造成误差传递;另外,现有的大多数模型在解码事件时,对事件数量的预测隐含在解码过程中,且只能按照预定义的事件顺序及预定义的角色顺序预测事件论元,使得先抽取的事件并没有考虑到后面抽取的事件。针对以上问题提出一种多任务联合的并行预测事件抽取框架。首先,使用预训练语言模型作为文档句子的编码器,检测文档中存在的事件类型,并使用结构化自注意力机制获取伪触发词特征,预测每种事件类型的事件数量;然后将伪触发词特征与候选论元特征进行交互,并行预测每个事件对应的事件论元,在大幅缩减模型训练时间的同时获得与基线模型相比更好的性能。最终事件抽取结果F1值为78%,事件类型检测子任务F1值为98.7%,事件数量预测子任务F1值为90.1%,实体识别子任务F1值为90.3%。

关键词：篇章级事件抽取多任务联合预训练语言模型结构化自注意力机制并行预测

The Instructional Design of Chinese Text Classification based on SVM 25th

在线全文

学校读者我要写书评

暂无评论

The Instructional Design of Chinese Text Classification base...

The Design of Digit Recognition Teaching Experiment Based on BP Neural Network

第25届中国控制与决策会议

作者： Sichao Wei Jianyi Guo Zhengtao Yu Peng Chen yantuan xian School of Information Engineering and Automation Kunming University of Science and Technology Intelligent Information Processing Key Laboratory Kunming University of Science and Technology

ISBN: (纸本)9781467355339

In order to resolve the comprehension difficulties of theory and implementation about Chinese text classification in " The principle and application of pattern recognition" curriculum for graduate students, this paper introduces the experiment of Chinese text classification into teaching practice. According to the text classification characteristics, we design the experiment scheme about Chinese text classification based on SVM, using word frequency statistics to extract feature and SVM classification algorithm, using vector space model to construct the feature space of text classification. So that readers can deeply understand and master the theoretical knowledge through the open the link, then expand on this basis

关键词： feature selection support vector machine algorithm text classification teaching experiment

来源： cnki会议评论

在线全文

cnki会议

学校读者我要写书评

暂无评论

The Design of Digit Recognition Teaching Experiment Based on...

Naxi-English Bilingual Word Alignment Based on Language Characteristics and Log-Linear Model

第25届中国控制与决策会议

作者： Haitao Yu Jianyi Guo Zhengtao Yu yantuan xian Peng Chen School of Information Engineering and Automation Kunming University of Science and Technology Intelligent Information Processing Key Laboratory Kunming University of Science and Technology

ISBN: (纸本)9781467355339

The teaching of pattern recognition course in postgraduates’ study is too theoretical to learn, and lack of teaching experiments. To solve these problems, this paper researches and explores on how to design digit recognition experiment in this *** propose an digit recognition teaching plan based on BP neural network. Firstly, we summarize the development and application of digit recognition, as well as the existing problems during the teaching activities at present;then, we use the case teaching method to explain the experimental procedures,and use the method of BP neural network to recognise digit, including the principle of digit recognition, the principle of image preprocessing and the design of neural network, and so on. With the aims of optimizing the original teaching form, stimulating students’ study interests, training their abilities of practice and scientific research , we introduce the design of experiment. Finally, help students to solve practical problems..

关键词： The practice of pattern recognition course BP neural network Digit recognition

来源： cnki会议评论

在线全文

cnki会议

学校读者我要写书评

暂无评论

China Communications 2012年第3期9卷 78-86页

作者： Yu Zhengtao xian yantuan Tian Wei Guo Jianyi Zhang Tao Kunming Univ Sci & Technol Sch Informat Engn & Automat Kunming 650550 Peoples R China

We propose a method that can achieve the Naxi-English bilingual word automatic alignment based on a log-linear *** method defines the different Naxi-English structural feature functions,which are English-Naxi interval switching function and Naxi-English bilingual word position transformation *** the manually labeled Naxi-English words alignment corpus,the parameters of the model are trained by using the minimum error,thus Naxi-English bilingual word alignment is achieved *** are conducted with IBM Model 3 as a benchmark,and the Naxi language constraints are *** final experiment results show that the proposed alignment method achieves very good results:the introduction of the language characteristic function can effectively improve the accuracy of the Naxi-English Bilingual Word Alignment.

关键词： word aligrmaent Naxi language English log-linear model interval switching function posi-tion transformation function

在线全文

学校读者我要写书评

暂无评论

结合类型感知注意力的少样本知识图谱补全

数据分析与知识发现 2023年第9期7卷 51-63页

作者：普祥和王红斌线岩团昆明理工大学信息工程与自动化学院昆明650500 昆明理工大学云南省人工智能重点实验室昆明650500 昆明理工大学云南省计算机技术应用重点实验室昆明650500

【目的】现有少样本知识图谱补全方法在处理复杂关系时不能很好地区分邻居重要性,导致实体预测性能不佳。考虑充分利用实体邻居信息,提高少样本知识图谱补全方法的性能。【方法】通过类型感知邻居编码器学习实体邻居中包含的隐含类型信... 详细信息

【目的】现有少样本知识图谱补全方法在处理复杂关系时不能很好地区分邻居重要性,导致实体预测性能不佳。考虑充分利用实体邻居信息,提高少样本知识图谱补全方法的性能。【方法】通过类型感知邻居编码器学习实体邻居中包含的隐含类型信息,得到类型感知注意力,增强实体表示;利用Transformer编码器捕获任务关系的不同含义;通过联合匹配原型网络聚合参考集得到参考集表示并进行实体预测。【结果】在NELL和Wiki两个公共数据集上进行实体预测任务,实验结果表明,MRR指标分别较Baseline方法提高了1.6和1.2个百分点。【局限】未对与实体相关性较低的邻居进行筛选,使类型感知注意力权重的分配受到噪声影响。【结论】本文方法能够通过学习更丰富的实体邻居信息来有效提高少样本知识图谱补全的性能。

关键词：少样本知识图谱补全实体预测类型感知注意力

The Teaching Experiment of Speech Recognition based on HMM 26th

在线全文

学校读者我要写书评

暂无评论

The Teaching Experiment of Speech Recognition based on HMM