CN116361476A - 一种基于插值法的知识图谱负样本合成方法 - Google Patents
一种基于插值法的知识图谱负样本合成方法 Download PDFInfo
- Publication number
- CN116361476A CN116361476A CN202211455256.2A CN202211455256A CN116361476A CN 116361476 A CN116361476 A CN 116361476A CN 202211455256 A CN202211455256 A CN 202211455256A CN 116361476 A CN116361476 A CN 116361476A
- Authority
- CN
- China
- Prior art keywords
- negative
- sample
- cand
- samples
- negative sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000001308 synthesis method Methods 0.000 title claims abstract description 9
- 238000012549 training Methods 0.000 claims abstract description 28
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 26
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 26
- 238000012216 screening Methods 0.000 claims abstract description 12
- 230000002194 synthesizing effect Effects 0.000 claims abstract description 9
- 238000009826 distribution Methods 0.000 claims description 16
- 238000010606 normalization Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 5
- 241000217776 Holocentridae Species 0.000 claims description 3
- 238000006467 substitution reaction Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 2
- 230000002349 favourable effect Effects 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
Abstract
Description
技术领域
本发明属于知识图谱嵌入领域,具体涉及一种基于插值法的知识图谱负样本合成方法。
背景技术
知识图谱(Knowledge Graph)是一种大规模的语义网络知识库,他采取符号化的表达方式,通过三元组的形式将知识存储于计算机中,因为知识图谱具有语义丰富、结构友好、易于理解等优势,所以近年来被广泛应用于态势感知、推荐系统、自然语言处理等领域。
虽然知识图谱具有明显优势,但是知识图谱中的知识是大量缺失的,为了补全知识图谱,目前最常用的技术是知识图谱嵌入。知识图谱嵌入是将知识图谱中的实体和关系嵌入到一个低维连续空间内,在方便计算的同时还保留知识图谱的结构信息。
知识图谱嵌入的训练过程中,需要提供正样本和负样本,使模型具备识别正、负样本的能力。正样本通常为现有的事实知识,负样本则是通过替换正样本中的头/尾实体而生成,该项技术称之为负采样技术。现有的负采样技术利用多种信息,在大量负样本候选集中进行筛选,得到有利于模型训练的强负样本,比如专利“一种知识图谱嵌入训练方法和相关装置”(CN202110013880.6)使用图谱的拓扑结构辅助筛选负样本。近年来有研究注意到mixup合成样本领域,比如专利“少标记半监督学习中的插值对比学习方法”(CN202210024335.1)利用mixup插值方法在嵌入空间中生成虚拟的正样本对,解决标签数据较少的问题。
发明内容
鉴于现有问题,本发明的目的在于提供一种基于插值法的知识图谱负样本合成方法,通过对算法等技术方案的改进,以解决上述技术问题。
本发明提供如下的技术方案:
一种基于插值法的知识图谱负样本合成方法,包括以下步骤:
S1:候选集筛选:从负样本中筛选负样本集合cand_il,作为mixup操作的候选集;S2:mixup样本合成:将集合cand_il中的负样本进行mixup混合得到cand_im,再将cand_im中的负样本和正样本进行mixup合成,得到强负样本cand_ik;S3:训练更新:将得到的负样本集合cand_il、cand_im、cand_ik再筛选一次得到cand_is,在模型训练中使用cand_is集合,更新强负样本集合/>
步骤S1包括以下步骤:
优选地,步骤S13通过以下公式计算相似度Ci:
步骤S2包括以下步骤:
S21:在负样本集合通过对相似度Ci进行归一化操作得到每个样本对应的概率P1i和候选集cand_il的多项概率分布,依据候选集的多项概率分布和每个样本对应的概率P1i对候选集cand_il进行两次抽样,对得到的两个样本/>和/>进行mixup合成操作;
S24:在负样本集合中,根据每个样本的相似度Cj,通过对的相似度Cj进行归一化操作计算每个样本对应的概率/> 和候选集cand_im的多项概率分布,通过概率P2j和候选集的多项概率分布对候选集cand_im进行一次抽样,将得到的一个负样本/>与正样本/>进行mixup合成操作;
计算mixup合成操作的结果,其中,αi为超参数,是样本/>和/>经过mixup合成之后的样本,||.||是对样本/>取L2正则化,其中,L2正则化公式为n为W的维度;步骤S24通过公式/> 计算概率P2j;通过公式:
步骤S23通过以下公式计算相似度Cj:
步骤S3包括以下步骤:
优选地,步骤S32通过公式:
计算权重Pi,其中,ε为超参数,s为cand_is中的样本总数。
优选地,当训练模型是基于平移距离的知识图谱嵌入模型时,损失函数为:
本发明的有益技术效果在于:
1.本发明提供的技术方案易于实现,运算效率高,不会增加原始嵌入模型的算法复杂度;
2.本发明提供的技术方案通过将mixup迁移到知识图谱嵌入领域,并进行双重mixup操作,增强了虚拟负样本的多样性;
3.本发明提供的技术方案可以挖掘出更多有利于模型训练的强负样本,提升知识图谱嵌入模型的整体性能;
4.本发明提供的技术方案可以叠加到任何已有的知识图谱嵌入模型中使用。
附图说明
图1是本发明实施例基于插值法的知识图谱负样本合成方法的流程示意图。
具体实施方式
下面对本发明的实施例作详细说明,下述的实施例在以本发明技术方案为前提下进行实施,给出了详细的实施方式和具体的操作过程,但本发明的保护范围不限于下述的实施例。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本申请保护的范围。
在本文中提及“实施例”意味着,结合实施例描述的特定特征、结构或特性可以包含在本申请的至少一个实施例中。在说明书中的各个位置出现该短语并不一定均是指相同的实施例,也不是与其它实施例互斥的独立的或备选的实施例。本领域技术人员显式地和隐式地理解的是,在不冲突的情况下,本文所描述的实施例可以与其它实施例相结合。
实施例1
参见图,在本发明实施例提供的基于插值法的知识图谱负样本合成方法,包括以下步骤:S1:候选集筛选:从负样本中筛选负样本集合cand_il,作为mixup操作的候选集;S2:mixup样本合成:将集合cand_il中的负样本进行mixup混合得到cand_im,再将cand_im中的负样本和正样本进行mixup合成,得到强负样本cand_ik;S3:训练更新:将得到的负样本集合cand_il、cand_im、cand_ik再筛选一次得到cand_is,在模型训练中使用cand_is集合,更新强负样本集合/>
步骤S1包括以下步骤:
步骤S13通过以下公式计算相似度Ci:
步骤S2包括以下步骤:
S21:在负样本集合通过对相似度Ci进行归一化操作得到每个样本对应的概率P1i和候选集cand_il的多项概率分布,依据候选集的多项概率分布和每个样本对应的概率P1i对候选集cand_il进行两次抽样,对得到的两个样本/>和/>进行mixup合成操作;
S24:在负样本集合中,根据每个样本的相似度Cj,通过对的相似度Cj进行归一化操作计算每个样本对应的概率/> 和候选集cand_im的多项概率分布,通过概率P2j和候选集的多项概率分布对候选集cand_im进行一次抽样,将得到的一个负样本/>与正样本/>进行mixup合成操作;
计算mixup合成操作的结果,其中,αi为超参数,是样本/>和/>经过mixup合成之后的样本,||·||是对样本/>取L2正则化,其中,L2正则化公式为n为W的维度;步骤S24通过公式/> 计算概率P2j;通过公式:
步骤S23通过以下公式计算相似度Cj:
步骤S3包括以下步骤:
步骤S32通过公式:
计算权重Pi,其中,ε为超参数,s为cand_is中的样本总数。
当训练模型是基于平移距离的知识图谱嵌入模型时,损失函数为:
实施例2
本发明另一优选实施例,在实施例1的基础上,在中文知识图谱中,对于某一个正样本为(江西,省份,中国),经过/>(假定k=5)嵌入表示为((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),通过从实体集合ε中任选头、尾实体进行随机替换生成大量负样本NSi,从NSi中任选n1(假定n1=5)个负样本:{((0.4,0.6,0.2,0.5,0.9),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),((0.5,0.3,0.1,0.7,0.2),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.2,0.3,0.9,0.5,0.4)),((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.6,0.5,0.9,0.7,0.5)),((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.3,0.1,0.2,0.3,0.5))},同时强负样本集合/> (假定h=3),将上述8个负样本混合后形成负样本集合cand_i:{((0.4,0.6,0.2,0.5,0.9),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),((0.5,0.3,0.1,0.7,0.2),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.2,0.3,0.9,0.5,0.4)),((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.6,0.5,0.9,0.7,0.5)),((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.3,0.1,0.2,0.3,0.5)),((0.4,0.2,0.9,0.4,0.6),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),((0.5,0.6,0.8,0.9,0.4),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.3,0.6,0.4,0.9,0.7))}。
计算cand_i中所有负样本和正样本之间的相似度Ci,得到相似度列表为{0.79,0.51,0.82,0.56,0.63,0.92,0.95,0.84},根据相似度降序排序,取top-l个样本(假定l=4),形成负样本集合cand_il:{((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.2,0.3,0.9,0.5,0.4)),((0.4,0.2,0.9,0.4,0.6),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),((0.5,0.6,0.8,0.9,0.4),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.3,0.6,0.4,0.9,0.7))}
通过对相似度Ci进行归一化操作得到每个样本对应的概率P1i和候选集cand_il的多项概率分布,依据候选集的多项概率分布和每个样本对应的概率P1i对候选集cand_il进行两次抽样和mixup合成操作,重复m次(假定m=4),挑选((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.2,0.3,0.9,0.5,0.4))和((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.3,0.6,0.4,0.9,0.7))进行mixup得到((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.25,0.45,0.65,0.7,0.55))(假定α1=0.5);挑选((0.4,0.2,0.9,0.4,0.6),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))和((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.3,0.6,0.4,0.9,0.7))进行mixup操作得到((0.25,0.35,0.75,0.35,0.4),(0.4,0.2,0.6,0.8,0.9),(0.4,0.35,0.5,0.9,0.7))(假定α2=0.5),挑选((0.5,0.6,0.8,0.9,0.4),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))和((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.3,0.6,0.4,0.9,0.7))进行mixup得到((0.3,0.55,0.7,0.6,0.3),(0.4,0.2,0.6,0.8,0.9),(0.4,0.35,0.5,0.9,0.7))(假定α3=0.5),挑选((0.4,0.2,0.9,0.4,0.6),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))和((0.5,0.6,0.8,0.9,0.4),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))进行mixup得到((0.45,0.4,0.85,0.65,0.5),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))(假定α4=0.5),最后得到cand_im:{((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.25,0.45,0.65,0.7,0.55)),((0.25,0.35,0.75,0.35,0.4),(0.4,0.2,0.6,0.8,0.9),(0.4,0.35,0.5,0.9,0.7)),((0.3,0.55,0.7,0.6,0.3),(0.4,0.2,0.6,0.8,0.9),(0.4,0.35,0.5,0.9,0.7)),((0.45,0.4,0.85,0.65,0.5),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))}。
计算cand_im中所有样本和正样本之间的相似度Cj,得到相似度列表为{0.89,0.98,0.87,0.92},通过对的相似度Cj进行归一化操作计算每个样本对应的概率和候选集cand_im的多项概率分布,依据候选集cand_im的多项概率分布和每个样本对应的概率P2j对候选集cand_im进行一次抽样并和正样本进行mixup操作,重复k次(假定k=4),挑选((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.25,0.45,0.65,0.7,0.55))样本((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))进行mixup得到((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.325,0.325,0.675,0.75,0.725))(假定β1=0.5),挑选((0.25,0.35,0.75,0.35,0.4),(0.4,0.2,0.6,0.8,0.9),(0.4,0.35,0.5,0.9,0.7))和正样本((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))进行mixup得到((0.175,0.425,0.675,0.325,0.3),(0.4,0.2,0.6,0.8,0.9),(0.45,0.225,0.55,0.9,0.7))(假定β2=0.5),挑选((0.3,0.55,0.7,0.6,0.3),(0.4,0.2,0.6,0.8,0.9),(0.4,0.35,0.5,0.9,0.7))正样本((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))进行mixup得到((0.2,0.525,0.65,0.45,0.25),(0.4,0.2,0.6,0.8,0.9),(0.45,0.225,0.55,0.9,0.7))(假定β3=0.5),挑选((0.45,0.4,0.85,0.65,0.5),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))和正样本((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))进行mixup得到((0.275,0.45,0.725,0.475,0.35),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))(假定β4=0.5),最后得到cand_ik:{((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.325,0.325,0.675,0.75,0.725)),((0.175,0.425,0.675,0.325,0.3),(0.4,0.2,0.6,0.8,0.9),(0.45,0.225,0.55,0.9,0.7)),((0.2,0.525,0.65,0.45,0.25),(0.4,0.2,0.6,0.8,0.9),(0.45,0.225,0.55,0.9,0.7)),((0.275,0.45,0.725,0.475,0.35),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))}。
混合负样本集合cand_il、cand_im和cand_ik得到cand_is:{((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.2,0.3,0.9,0.5,0.4)),((0.4,0.2,0.9,0.4,0.6),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),((0.5,0.6,0.8,0.9,0.4),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.3,0.6,0.4,0.9,0.7)),((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.25,0.45,0.65,0.7,0.55)),((0.25,0.35,0.75,0.35,0.4),(0.4,0.2,0.6,0.8,0.9),(0.4,0.35,0.5,0.9,0.7)),((0.3,0.55,0.7,0.6,0.3),(0.4,0.2,0.6,0.8,0.9),(0.4,0.35,0.5,0.9,0.7)),((0.45,0.4,0.85,0.65,0.5),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7)),((0.1,0.5,0.6,0.3,0.2),(0.4,0.2,0.6,0.8,0.9),(0.325,0.325,0.675,0.75,0.725)),((0.175,0.425,0.675,0.325,0.3),(0.4,0.2,0.6,0.8,0.9),(0.45,0.225,0.55,0.9,0.7)),((0.2,0.525,0.65,0.45,0.25),(0.4,0.2,0.6,0.8,0.9),(0.45,0.225,0.55,0.9,0.7)),((0.275,0.45,0.725,0.475,0.35),(0.4,0.2,0.6,0.8,0.9),(0.5,0.1,0.6,0.9,0.7))},根据上一轮的嵌入模型计算权重Pi列表为:{0.95,0.65,0.574,0.85,0.42,0.285,0.65,0.21,0.98,0.356,0.36,0.6},将权重降序排序,取top-h(假定h=3)个负样本,更新强负样本集合
本发明上述实施例提供的方法,易于实现,运算速度快,不增加原始嵌入模型的复杂度;能增强虚拟负样本的多样性,提升知识图谱嵌入模型的性能,易于叠加到已有的知识图谱嵌入模型中。
以上详细描述了本发明的较佳具体实施例。应当理解,本领域的普通技术无需创造性劳动就可以根据本发明的构思做出诸多修改和变化。因此,凡本技术领域中技术人员依本发明的构思在现有技术的基础上通过逻辑分析、推理或者有限的试验可以得到的技术方案,皆应在由权利要求书所确定的保护范围内。
Claims (9)
2.根据权利要求1所述的基于插值法的知识图谱负样本合成方法,其特征在于,所述步骤S1包括以下步骤:
4.根据权利要求3所述的基于插值法的知识图谱负样本合成方法,其特征在于,所述步骤S2包括以下步骤:
S21:在负样本集合通过对相似度Ci进行归一化操作得到每个样本对应的概率P1i和候选集cand_il的多项概率分布,依据候选集的多项概率分布和每个样本对应的概率P1i对候选集cand_il进行两次抽样,对得到的两个样本/>和/>进行mixup合成操作;
S24:在负样本集合中,根据每个样本的相似度Cj,通过对的相似度Cj进行归一化操作计算每个样本对应的概率/> 和候选集cand_im的多项概率分布,通过概率P2j和候选集的多项概率分布对候选集cand_im进行一次抽样,将得到的一个负样本/>与正样本/>进行mixup合成操作;
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211455256.2A CN116361476B (zh) | 2022-11-21 | 2022-11-21 | 一种基于插值法的知识图谱负样本合成方法 |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211455256.2A CN116361476B (zh) | 2022-11-21 | 2022-11-21 | 一种基于插值法的知识图谱负样本合成方法 |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116361476A true CN116361476A (zh) | 2023-06-30 |
CN116361476B CN116361476B (zh) | 2024-05-17 |
Family
ID=86915203
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211455256.2A Active CN116361476B (zh) | 2022-11-21 | 2022-11-21 | 一种基于插值法的知识图谱负样本合成方法 |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116361476B (zh) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112182245A (zh) * | 2020-09-28 | 2021-01-05 | 中国科学院计算技术研究所 | 一种知识图谱嵌入模型的训练方法、系统和电子设备 |
US20210224690A1 (en) * | 2020-01-21 | 2021-07-22 | Royal Bank Of Canada | System and method for out-of-sample representation learning |
CN115048538A (zh) * | 2022-08-04 | 2022-09-13 | 中国科学技术大学 | 基于关系增强负采样的多模态知识图谱补全方法与系统 |
-
2022
- 2022-11-21 CN CN202211455256.2A patent/CN116361476B/zh active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20210224690A1 (en) * | 2020-01-21 | 2021-07-22 | Royal Bank Of Canada | System and method for out-of-sample representation learning |
CN112182245A (zh) * | 2020-09-28 | 2021-01-05 | 中国科学院计算技术研究所 | 一种知识图谱嵌入模型的训练方法、系统和电子设备 |
CN115048538A (zh) * | 2022-08-04 | 2022-09-13 | 中国科学技术大学 | 基于关系增强负采样的多模态知识图谱补全方法与系统 |
Non-Patent Citations (2)
Title |
---|
TIROSHAN MADUSHANKA 等: "MDNCaching: A Strategy to Generate Quality Negatives for Knowledge Graph Embedding", SPRINGER NATURE SWITZERLAND AG 2022, 30 June 2022 (2022-06-30), pages 877 - 888 * |
雷景生 等: "基于上下文语义增强的实体关系联合抽取", 计算机应用, vol. 43, no. 5, 30 September 2022 (2022-09-30), pages 1438 - 1444 * |
Also Published As
Publication number | Publication date |
---|---|
CN116361476B (zh) | 2024-05-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110413986B (zh) | 一种改进词向量模型的文本聚类多文档自动摘要方法及系统 | |
CN109325229B (zh) | 一种利用语义信息计算文本相似度的方法 | |
Shi et al. | Deep adaptively-enhanced hashing with discriminative similarity guidance for unsupervised cross-modal retrieval | |
CN108460012A (zh) | 一种基于gru-crf的命名实体识别方法 | |
CN110826303A (zh) | 一种基于弱监督学习的联合信息抽取方法 | |
Liu et al. | Optimization-based key frame extraction for motion capture animation | |
CN110263174A (zh) | —基于焦点关注的主题类别分析方法 | |
WO2021026044A1 (en) | Framework for learning to transfer learn | |
CN113220865A (zh) | 一种文本相似词汇检索方法、系统、介质及电子设备 | |
CN106650820A (zh) | 一种手写电气元器件符号与标准电气元器件符号的匹配识别方法 | |
CN117216578A (zh) | 基于元学习的可自定义标签深度学习模型构建方法及系统 | |
CN116361476A (zh) | 一种基于插值法的知识图谱负样本合成方法 | |
CN108256030A (zh) | 一种基于本体的密度自适应概念语义相似度计算方法 | |
Cihan Camgoz et al. | Particle filter based probabilistic forced alignment for continuous gesture recognition | |
Wu et al. | Recognition of pear leaf disease under complex background based on DBPNet and modified mobilenetV2 | |
Feng et al. | Prototypical networks relation classification model based on entity convolution | |
Xu et al. | A sophisticated offline network developed for recognizing handwritten Chinese character efficiently | |
CN110597982A (zh) | 一种基于词共现网络的短文本主题聚类算法 | |
Wu et al. | Active 3-D shape cosegmentation with graph convolutional networks | |
CN113011519A (zh) | 一种多尺度分类数据挖掘方法 | |
Hu et al. | VIGraph: Self-supervised Learning for Class-Imbalanced Node Classification | |
Peng et al. | Named entity recognition based on reinforcement learning and adversarial training | |
CN109146058A (zh) | 具有变换不变能力且表达一致的卷积神经网络 | |
Zhang et al. | Leaf Cultivar Identification via Prototype-enhanced Learning | |
CN109101570A (zh) | 一种基于图摘要的图模式挖掘方法 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |