融合噪声过滤的超关系知识图谱补全方法

doi:10.11772/j.issn.1001-9081.2024060792

《计算机应用》唯一官方网站 ›› 2025, Vol. 45 ›› Issue (6): 1817-1826.DOI: 10.11772/j.issn.1001-9081.2024060792

• 人工智能 • 上一篇

融合噪声过滤的超关系知识图谱补全方法

刘爽(), 刘大庆, 孟佳娜, 赵迪

大连民族大学计算机科学与工程学院，辽宁大连 116600

收稿日期:2024-06-14 修回日期:2024-08-30 接受日期:2024-09-05 发布日期:2024-09-25 出版日期:2025-06-10
通讯作者: 刘爽
作者简介:刘爽（1977—），女，辽宁锦州人，教授，博士，CCF会员，主要研究方向：知识图谱、深度学习 dlnuliushuang@qq.com
刘大庆（1999—），男，河北张家口人，硕士研究生，主要研究方向：知识图谱补全、链接预测
孟佳娜（1972—），女，吉林四平人，教授，博士，CCF会员，主要研究方向：机器学习、文本挖掘
赵迪（1991—），男，吉林四平人，讲师，博士，主要研究方向：数据挖掘、自然语言处理。
基金资助:
2023年度教育部人文社会科学研究规划基金资助项目(23YJA860010)

Hyper-relational knowledge graph completion method fusing noise filtering

Shuang LIU(), Daqing LIU, Jiana MENG, Di ZHAO

Computer Science and Engineering College，Dalian Minzu University，Dalian Liaoning 116600，China

Received:2024-06-14 Revised:2024-08-30 Accepted:2024-09-05 Online:2024-09-25 Published:2025-06-10
Contact: Shuang LIU
About author:LIU Shuang， born in 1977， Ph. D.， professor. Her research interests include knowledge graph， deep learning.
LIU Daqing， born in 1999， M. S. candidate. His research interests include knowledge graph completion， link prediction.
MENG Jiana， born in 1972， Ph. D.， professor. Her research interests include machine learning， text mining.
ZHAO Di， born in 1991， Ph. D.， lecturer. His research interests include data mining， natural language processing.
Supported by:
2023 Humanities and Social Sciences Research and Planning Fund of Ministry of Education(23YJA860010)

摘要/Abstract

摘要：

针对超关系知识图谱中限定符会为主三元组引入无关噪声的问题，提出一种融合噪声过滤的超关系知识图谱补全方法（HRNF）。首先，为了有效增强超关系事实，构建特征增强模块；同时，利用卷积神经网络（CNN）提取普通三元组特征，并通过异构图神经网络（HGNN）捕获超关系事实中的复杂关系特征；其次，融合这2种特征，利用普通三元组的稳定性与可靠性增强超关系事实中主三元组的信息，减少限定符引入噪声的影响；再次，为了更准确地融合特征表示，构建相关性感知模块；同时，利用图注意力网络（GATv2），通过动态学习不同节点间的权重更新增强后的特征表示；继次，为了捕获复杂的语义信息，构建语义增强模块；最后利用Transformer模型，通过自注意力机制捕获序列中任意2个元素之间的依赖关系，从而生成最终的预测序列。为了验证HRNF的有效性，在2个常用的数据集Wikipeople和JF17K上进行广泛的实验。结果表明，相较于基线方法中较优的GRAN（GRAph-based N-ary relational learning），在预测主三元组实体时，HRNF在Wikipeople数据集上的平均倒数排名（MRR）、Hits@1和Hits@10分别提升了0.6、1.1和1.8个百分点，在JF17K数据集上的MRR、Hits@1和Hits@10分别提升了0.5、0.7和2.9个百分点。以上这些显著提升证明了HRNF在处理超关系知识图谱补全任务中可以有效地缓解限定符带来的噪声问题。

关键词: 噪声过滤, 限定符, 超关系事实, 超关系知识图谱补全, 普通三元组

Abstract:

Aiming at the problem that qualifiers in the hyper-relational knowledge graph will introduce irrelevant noise into the main triple， a Hyper-Relational knowledge graph completion method fusing Noise Filtering （HRNF） was proposed. Firstly， a feature enhancement module was constructed in order to enhance the hyper-relational facts effectively. At the same time， Convolutional Neural Network （CNN） was utilized to extract the ordinary triple features， and complex relational features in the hyper-relational fact were captured by Heterogeneous Graph Neural Network （HGNN）. Secondly， these two features were fused to enhance information of the main triple in the hyper-relational fact by utilizing stability and reliability of the ordinary triple， so as to reduce the effect of noise introduced by qualifiers. Thirdly， a relevance-aware module was constructed to fuse the feature representations more accurately. At the same time， Graph ATtention network version Two（GATv2） was utilized to update the enhanced feature representation by learning weights among different nodes dynamically. Fourthly， a semantic enhancement module was constructed to capture complex semantic information. Finally， Transformer model was utilized to generate the final predicted sequence by capturing the dependency between any two elements in the sequence through self-attention mechanism. To validate the effectiveness of HRNF， extensive experiments were conducted on two commonly used datasets， Wikipeople and JF17K. The results show that when predicting main triple entities， compared to the optimal GRAN （GRAph-based N-ary relational learning） of the baseline methods， the Mean Reciprocal Rank （MRR）， Hits@1， and Hits@10 of HRNF are improved by 0.6， 1.1， and 1.8 percentage points， respectively， on Wikipeople dataset， and the MRR， Hits@1， and Hits@10 of HRNF are improved by 0.5， 0.7， and 2.9 percentage points， respectively， on JF17K dataset. The above significant improvements prove that in dealing with task of hyper-relational knowledge graph completion， HRNF can reduce the noise problem brought by qualifiers effectively.

Key words: noise filtering, qualifier, hyper-relational fact, hyper-relational knowledge graph completion, ordinary triple

中图分类号:

TP391.1

刘爽, 刘大庆, 孟佳娜, 赵迪. 融合噪声过滤的超关系知识图谱补全方法[J]. 计算机应用, 2025, 45(6): 1817-1826.

Shuang LIU, Daqing LIU, Jiana MENG, Di ZHAO. Hyper-relational knowledge graph completion method fusing noise filtering[J]. Journal of Computer Applications, 2025, 45(6): 1817-1826.

图/表 15

图1 超关系事实

Fig. 1 Hyper-relational facts

图2 本文方法的总体框架

Fig. 2 Overall framework of proposed method

表1 数据集详情

Tab. 1 Dataset details

实体

数

关系数

超关系数

最大

属性值

表2 超参数的最佳设置

Tab. 2 Best setting of hyperparameters

超参数	Wikipeople参数值	JF17K参数值
训练轮数	300	300
学习率	$5 × 10 - 4$	$5 × 10 - 4$
嵌入维度	256	256
图表示丢弃值	0.1	0.2
图表示注意力头数	4	4
语义丢弃值	0.1	0.2
语义注意力头数	4	4
HGNN激活函数	eLU	eLU
解码器激活函数	GeLU	GeLU
隐藏层大小	256	256
批次大小	1 024	1 024
权重衰减	0.01	0.01
实体软标签	0.2	0.9
关系软标签	0.1	0.0

表2 超参数的最佳设置

Tab. 2 Best setting of hyperparameters

超参数	Wikipeople参数值	JF17K参数值
训练轮数	300	300
学习率	$5 × 10 - 4$	$5 × 10 - 4$
嵌入维度	256	256
图表示丢弃值	0.1	0.2
图表示注意力头数	4	4
语义丢弃值	0.1	0.2
语义注意力头数	4	4
HGNN激活函数	eLU	eLU
解码器激活函数	GeLU	GeLU
隐藏层大小	256	256
批次大小	1 024	1 024
权重衰减	0.01	0.01
实体软标签	0.2	0.9
关系软标签	0.1	0.0

表3 链接预测结果

Tab. 3 Link prediction results

预测位置	方法	Wikipeople						JF17K
		主三元组			所有实体			主三元组			所有实体
		MRR	Hits@1	Hits@10	MRR	Hits@1	Hits@10	MRR	Hits@1	Hits@10	MRR	Hits@1	Hits@10
实体预测	m-TransH	0.063	0.063	0.300	—	—	—	0.206	0.206	0.462	0.102	0.069	0.168
	RAE	0.058	0.058	0.306	0.172	0.102	0.320	0.215	0.215	0.466	0.310	0.219	0.504
	NaLP	0.408	0.331	0.546	0.338	0.272	0.466	0.221	0.165	0.331	0.366	0.290	0.516
	HINGE	0.342	0.272	0.463	0.350	0.282	0.467	0.431	0.342	0.611	0.517	0.436	0.675
	StarE	0.491	0.398	0.592	0.378	0.265	0.542	0.574	0.496	0.725	0.542	0.454	0.685
	Hy-Transformer	0.501	0.426	0.634	—	—	—	0.582	0.501	0.742	—	—	—
	GRAN	0.503	0.438	0.620	0.479	0.410	0.604	0.617	0.539	0.770	0.656	0.582	0.799
	QUAD	0.497	0.431	0.617	—	—	—	0.596	0.519	0.751	—	—	—
	HyperFormer	—	—	—	0.473	0.361	0.646	—	—	—	0.664	0.607	0.787
	HRNF	0.509	0.449	0.638	0.477	0.424	0.649	0.622	0.546	0.799	0.666	0.593	0.813
关系预测	NaLP	0.482	0.320	0.482	0.735	0.595	0.938	0.639	0.547	0.822	0.825	0.762	0.927
	HINGE	—	—	—	0.765	0.686	0.900	—	—	—	0.861	0.832	0.910
	StarE	—	—	—	0.378	0.265	0.542	—	—	—	0.901	0.884	0.963
	GRAN	0.957	0.942	0.976	0.960	0.946	0.977	0.992	0.988	0.988	0.996	0.993	0.999
	HRNF	0.960	0.945	0.976	0.953	0.945	0.977	0.993	0.991	0.996	0.994	0.994	0.999

图3 Wikipeople数据集上的特征增强分析

Fig. 3 Feature enhancement analysis on Wikipeople dataset

图4 JF17K数据集上的特征增强分析

Fig. 4 Feature enhancement analysis on JF17K dataset

图5 学习率敏感性分析

Fig. 5 Learning rate sensitivity analysis

图6 嵌入维度敏感性分析

Fig. 6 Embedded dimension sensitivity analysis

图7 多头注意力敏感性分析

Fig. 7 Multi-head attention sensitivity analysis

表4 消融实验结果

Tab. 4 Ablation experimental results

方法	Wikipeople						JF17K
	主三元组			所有实体关系			主三元组			所有实体关系
	MRR	Hits@1	Hits@10	MRR	Hits@1	Hits@10	MRR	Hits@1	Hits@10	MRR	Hits@1	Hits@10
HRNF-C	0.482	0.424	0.619	0.466	0.403	0.624	0.591	0.536	0.769	0.641	0.563	0.772
HRNF-H	0.491	0.439	0.612	0.469	0.411	0.637	0.610	0.543	0.779	0.653	0.576	0.781
HRNF-R	0.499	0.434	0.626	0.471	0.417	0.635	0.613	0.539	0.788	0.649	0.582	0.792
HRNF-T	0.471	0.422	0.611	0.436	0.394	0.614	0.603	0.512	0.762	0.633	0.545	0.740

图8 消融实验中的多头注意力敏感性分析

Fig. 8 Multi-head attention sensitivity analysis in ablation study

图9 超图表示消融分析

Fig.9 Hypergraph representation ablation analysis

图10 训练损失

Fig. 10 Training loss

图11 测试损失

Fig. 11 Testing loss

参考文献 35

1	张文豪，徐贞顺，刘纳，等. 知识图谱补全方法研究综述［J］. 计算机工程与应用， 2024， 60（12）：61-73.
	ZHANG W H， XU Z S， LIU N， et al. Overview of knowledge graph completion methods［J］. Computer Engineering and Applications， 2024， 60（12）：61-73.
2	BORDES A， USUNIER N， GARCIA-DURÁN A， et al. Translating embeddings for modeling multi-relational data［C］// Proceedings of the 27th International Conference on Neural Information Processing Systems — Volume 2. Red Hook： Curran Associates Inc.， 2013： 2787-2795.
3	NICKEL M， TRESP V， KRIEGEL H P. A three-way model for collective learning on multi-relational data［C］// Proceedings of the 28th International Conference on Machine Learning. Madison， WI： Omnipress， 2011： 809-816.
4	DETTMERS T， MINERVINI P， STENETORP P， et al. Convolutional 2D knowledge graph embeddings ［C］// Proceedings of the 32nd AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2018： 1811-1818.
5	GUAN S， JIAN X， GUO J， et al. Link prediction on n-ary relational data based on relatedness evaluation［J］. IEEE Transactions on Knowledge and Data Engineering， 2023， 35（1）： 672-685.
6	WEN J， LI J， MAO Y， et al. On the representation and embedding of knowledge bases beyond binary relations［C］// Proceedings of the 25th International Joint Conference on Artificial Intelligence. California： ijcai.org， 2016： 1300-1307.
7	ZHANG R， LI J， MEI J， et al. Scalable instance reconstruction in knowledge bases via relatedness affiliated embedding［C］// Proceedings of the 2018 World Wide Web Conference. Republic and Canton of Geneva： International World Wide Web Conferences Steering Committee， 2018： 1185-1194.
8	ROSSO P， YANG D， CUDRÉ-MAUROUX P. Beyond triplets： hyper-relational knowledge graph embedding for link prediction［C］// Proceedings of the Web Conference 2020. New York： ACM， 2020： 1885-1896.
9	GALKIN M， TRIVEDI P， MAHESHWARI G， et al. Message passing for hyper-relational knowledge graphs［C］// Proceedings of the 2020 Empirical Methods in Natural Language Processing. Stroudsburg： ACL， 2020： 7346-7359.
10	HU Z， GUTIÉRREZ-BASULTO V， XIANG Z， et al. HyperFormer： enhancing entity and relation interaction for hyper-relational knowledge graph completion［C］// Proceedings of the 32nd ACM International Conference on Information and Knowledge Management. New York： ACM， 2023： 803-812.
11	LUO H， E H， TAN L， et al. DHGE： dual-view hyper-relational knowledge graph embedding for link prediction and entity typing［C］// Proceedings of the 37th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2023： 6467-6474.
12	WANG C， LI X， GAN T， et al. Hyper-relational knowledge graph embedding based on type constraints ［C］// Proceedings of the 16th International Conference on Advanced Computer Theory and Engineering. Piscataway： IEEE， 2023： 54-58.
13	LUO H， YANG Y E H， et al. HAHE： hierarchical attention for hyper-relational knowledge graphs in global and local level［C］// Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics （Volume 1： Long Papers）. Stroudsburg： ACL， 2023： 8095-8107.
14	CHUNG C， LEE J， WHANG J. Representation learning on hyper-relational and numeric knowledge graphs with Transformers ［C］// Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York： ACM， 2023： 310-322.
15	FENG Y， YOU H， ZHANG Z， et al. Hypergraph neural networks［C］// Proceedings of the 33rd AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2019： 3558-3565.
16	BRODY S， ALON U， YAHAV E. How attentive are graph attention networks？［EB/OL］. ［2024-04-23］..
17	WANG Z， ZHANG J， FENG J， et al. Knowledge graph embedding by translating on hyperplanes［C］// Proceedings of the 28th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2014： 1112-1119.
18	JI G， HE S， XU L， et al. Knowledge graph embedding via dynamic mapping matrix［C］// Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing （Volume 1： Long Papers）. Stroudsburg： ACL， 2015： 687-696.
19	LIN Y， LIU Z， SUN M， et al. Learning entity and relation embeddings for knowledge graph completion［C］// Proceedings of the 29th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2015： 2181-2187.
20	YANG B， YIH W T， HE X， et al. Embedding entities and relations for learning and inference in knowledge bases［EB/OL］. ［2024-04-23］..
21	TROUILLON T， WELBL J， RIEDEL S， et al. Complex embeddings for simple link prediction［C］// Proceedings of the 33rd International Conference on Machine Learning. New York： JMLR.org， 2016， 2071-2080.
22	NGUYEN D Q， TUDINH T D， NGUYEN D Q， et al. A novel embedding model for knowledge base completion based on convolutional neural network［C］// Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics： Human Language Technologies， Volume 2 （Short Papers）. Stroudsburg： ACL， 2018： 327-333.
23	SUN Z， DENG Z H， NIE J Y， et al. RotatE： knowledge graph embedding by relational rotation in complex space［EB/OL］. ［2024-04-23］..
24	SONG T， LUO J， HUANG L. Rot-Pro： modeling transitivity by projection in knowledge graph embedding［C］// Proceedings of the 35th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2021： 24695-24706.
25	JIANG T， LIU T， GE T， et al. Encoding temporal information for time-aware link prediction［C］// Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing. Stroudsburg： ACL， 2016： 2350-2354.
26	DASGUPTA S S， RAY S N， TALUKDAR P. HyTE： hyperplane-based temporally aware knowledge graph embedding［C］// Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Stroudsburg： ACL， 2018： 2001-2011.
27	GOEL R， KAZEMI S M， BRUBAKER M， et al. Diachronic embedding for temporal knowledge graph completion［C］// Proceedings of the 34th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2020： 3988-3995.
28	JAIN P， RATHI S， CHAKRABARTI S， et al. Temporal knowledge base completion： new algorithms and evaluation protocols［C］// Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing. Stroudsburg： ACL， 2020： 3733-3747.
29	XU C， NAYYERI M， ALKHOURY F， et al. TeRo： a time-aware knowledge graph embedding via temporal rotation［C］// Proceedings of the 28th International Conference on Computational Linguistics. ［S.l.］： International Committee on Computational Linguistics， 2020： 1583-1593.
30	SHOMER H， JIN W， LI J， et al. Learning representations for hyper-relational knowledge graphs［C］// ASONAM ’23：Proceedings of the International Conference on Advances in Social Networks Analysis and Mining. New York： ACM， 2023： 253-257.
31	LOSHCHILOV I， HUTTER F. Decoupled weight decay regularization［EB/OL］. ［2024-04-25］..
32	VRANDEČIĆ D， KRÖTZSCH M. Wikidata： a free collaborative knowledgebase ［J］. Communications of the ACM， 2014， 57（10）： 78-85.
33	BOLLACKER K， EVANS C， PARITOSH P， et al. Freebase： a collaboratively created graph database for structuring human knowledge［C］// Proceedings of the 2008 ACM SIGMOD International Conference on Management of Data. New York： ACM， 2008： 1247-1250.
34	YU D， YANG Y. Improving hyper-relational knowledge graph completion［EB/OL］. ［2024-04-28］..
35	WANG Q， WANG H， LYU Y， et al. Link prediction on n-ary relational facts： a graph-based approach［C］// Findings of the Association for Computational Linguistics： ACL-IJCNLP 2021. Stroudsburg： ACL， 2021： 396-407.

[1]	张增辉, 姜高霞, 王文剑. 基于动态概率抽样的标签噪声过滤方法[J]. 《计算机应用》唯一官方网站, 2021, 41(12): 3485-3491.
[2]	张增辉, 姜高霞, 王文剑. 基于局部概率抽样的标签噪声过滤方法[J]. 计算机应用, 2021, 41(1): 67-73.
[3]	张岩闫德勤郑宏亮. 非球形分布数据集的去噪方法[J]. 计算机应用, 2011, 31(10): 2786-2789.

融合噪声过滤的超关系知识图谱补全方法

Hyper-relational knowledge graph completion method fusing noise filtering

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 15

参考文献 35

相关文章 3

编辑推荐

Metrics