语义区分和子图整合驱动的知识图谱表示学习方法

Knowledge Graph Representation Learning Driven by Semantic Discrimination and Subgraph Integration

摘要

针对现有基于图神经网络的知识图谱表示学习方法难以区分实体的多样化语义，以及如何 在嵌入学习过程中对所有关系进行均质化处理的问题，提出了一种新颖的基于语义区分和子图整 合知识图谱表示学习方法，通过层次化建模，分阶段挖掘图谱结构信息以学习实体表示。语义区分 模块通过构建多个关系感知子图并整合其语义信息，从而全面捕获实体属性。子图整合模块则通 过评估不同子图的重要性，实施选择性邻域聚合以保留具有区分度的语义特征。此外，采用编码器 对知识图谱三元组进行建模并执行链接预测任务，并引入交叉熵损失函数训练模型，以充分整合语 义与结构信息。实验结果表明：所提方法在3个标准数据集上相较于现有最优模型实现了全面性 能提升。在FB15K-237数据集上，平均倒数排名(M_RR)、命中率Hits@3和Hits@10(正确答案出 现在预测排名前3、前10位的比例)分别提升了2.42%、3.90%和1.44%；在YAGO3-10数据集 上，M_RR、Hits@3和Hits@10分别提升了2.17%、4.81%和2.01%；在WN18RR数据集上，M_RR、 Hits@3和Hits@10分别提升了0.61%、0.59%和0.86%，有效验证了所提方法的优越性。

Abstract

To address the limitations of existing graph neural network-based knowledge graph representation learning methods in distinguishing diverse entity semantics and handling relations homogeneously during embedding learning, a novel knowledge graph representation learning method based on semantic discrimination and subgraph integration is proposed. Hierarchical modeling is employed to learn entity representations by progressively mining structural information from the graph. In the semantic discrimination module, multiple relation-aware subgraphs are constructed and their semantic information is integrated to capture comprehensive entity attributes. In the subgraph integration module, the importance of different subgraphs is evaluated, and selective neighborhood aggregation is performed to preserve discriminative semantic features. Additionally, knowledge graph triples are modeled via an encoder for link prediction tasks, and a cross-entropy loss function is introduced during model training to effectively integrate semantic and structural information. Experimental results on three standard datasets demonstrate that the proposed method yields comprehensive performance improvements over current state-of-the-art models. Specifically, on the FB15K-237 dataset, improvements of 2.42%, 3.90%, and 1.44% were achieved in mean reciprocal rank (MRR), Hits@3, and Hits@10 (denoting the proportion of queries with the correct answer in the topk predictions (k=3, 10)) respectively. On the YAGO3-10 dataset, the MRR, Hits@3, and Hits@10 were improved by 2.17%, 4.81%, and 2.01%, respectively. On the WN18RR dataset, increases of 0.61%, 0.59%, and 0.86% were observed in the corresponding metrics. The superiority of the proposed method is thus validated.

关键词

知识图谱 / 表示学习 / 图神经网络 / 链接预测

Key words

knowledge graph / representation learning / graph neural networks / link prediction

参考文献

[1] Liang Ke, Meng Lingyuan, Liu Meng, et al. A survey of knowledge graph reasoning on graph types: static, dynamic, and multi—modal[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(12): 9456-9478.

[2] 朱可，秦涛，朱奎宇，等 . 采用消息聚合的演化感知时序知识图谱推理方法[J].西安交通大学学报，2025，59(11): 187-197.

[3] Zhu Ke, Qin Tao, Zhu Kuiyu, et al. Evolution—aware temporal knowledge graph reasoning method based on message aggregation[J].Journal of Xi’an Jiaotong University, 2025, 59(11): 187-197.

[4] Shang Bin, Zhao Yinliang, Liu Jun, et al. LAFA: multimodal knowledge graph completion with link aware fusion and aggregation[C]//Proceedings of the Thirty—Eighth AAAI Conference on Artificial Intelligence and Thirty—Sixth Conference on Innovative Applications of Artificial Intelligence and Fourteenth Symposium on Educational Advances in Artificial Intelligence. Palo Alto, CA, USA: AAAI Press, 2024: 8957-8965.

[5] Trouillon T, Welbl J, Riedel S, et al. Complex embeddings for simple link prediction[C]//Proceedings of the 33rd International Conference on International Conference on Machine Learning. Chia Laguna Resort, Sardinia, Italy: PMLR, 2016: 2071-2080.

[6] 吴信东，黄满宗，卜晨阳 . BEKO: 大语言模型与知识图谱的双向增强[J].计算机学报，2025，48(7): 1572-1588.

[7] Wu Xindong, Huang Manzong, Bu Chenyang . BEKO: bidirectional enhancement with a knowledge ocean for LLMs and KGs[J].Chinese Journal of Computers, 2025, 48(7): 1572-1588.

[8] Shang Bin, Zhao Yinliang, Wang Di, et al. Relation—aware multi—positive contrastive knowledge graph completion with embedding dimension scaling[C]//Proceedings of the 46th international ACM SIGIR conference on research and development in information retrieval. New York, USA: Association for Computing Machinery, 2023: 878-888.

[9] 郑修林，周鹏，李培培，等 . 知识图谱补全技术及应用[J].软件学报，2025，36(12): 5599-5628.

[10] Zheng Xiulin, Zhou Peng, Li Peipei, et al. Knowledge graph completion: techniques and applications[J].Journal of Software, 2025, 36(12): 5599-5628.

[11] Mahdisoltani F, Biega J, Suchanek F M . YAGO3: a knowledge base from multilingual wikipedias[EB/OL]. (2015—01—04)[2025—12—05].https://www.semantic—scholar.org/paper/YAGO3:—A—Knowledge—Base—from—Multilingual—Mahdisoltani—Biega/6c5b5adc3830ac45bf1d764603b1b71e5f729616.

[12] 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, USA: ACM, 2008: 1247-1250.

[13] Dettmers T, Minervini P, Stenetorp P, et al. Convolutional 2D knowledge graph embeddings[C]//Proceedings of the Thirty—Second AAAI Conference on Artificial Intelligence and Thirtieth Innovative Applications of Artificial Intelligence Conference and Eighth AAAI Symposium on Educational Advances in Artificial Intelligence. Palo Alto, CA, USA: AAAI Press, 2018: 1811-1818.

[14] Sun Zhiqing, Deng Zhihong, Nie Jianyun, et al. RotatE: knowledge graph embedding by relational rotation in complex space[C]//ICLR 2019 Conference. New York, USA: ICLR, 2019: 1-18.

[15] Shang Bin, Zhao Yinliang, Liu Jun . Learnable convolutional attention network for knowledge graph completion[J].Knowledge—Based Systems, 2024, 285: 111360.

[16] Bordes A, Usunier N, Garcia—duran A, et al. Translating embeddings for modeling multi—relational data[C]//Proceedings of the 27th International Conference on Neural Information Processing Systems. Red Hook, NY, USA: Curran Associates Inc., 2013: 2787-2795.

[17] Shi Fobo, Li Duantengchuan, Wang Xiaoguang, et al. TGformer: a graph transformer framework for knowledge graph embedding[J].IEEE Transactions on Knowledge and Data Engineering, 2025, 37(1): 526-541.

[18] Vashishth S, Sanyal S, Nitin V, et al. Composition—based multi—relational graph convolutional networks[C]//ICLR2020 Conference. New York, USA: ICLR, 2020: 1-16.

[19] Zhang Zhanqiu, Wang Jie, Ye Jieping, et al. Rethinking graph convolutional networks in knowledge graph completion[C]//Proceedings of the ACM Web Conference 2022. New York, USA: ACM, 2022: 798-807.

[20] Li Ren, Cao Yanan, Zhu Qiannan, et al. How does knowledge graph embedding extrapolate to unseen data: a semantic evidence view[C]//Proceedings of the Thirty—Sixth AAAI Conference on Artificial Intelligence and Thirty—Fourth Conference on Innovative Applications of Artificial Intelligence and the Twelveth Symposium on Educational Advances in Artificial Intelligence. Palo Alto, CA, USA: AAAI Press, 2022: 5781-5791.

[21] Dai Guoquan, Wang Xizhao, Zou Xiaoying, et al. MRGAT: multi—relational graph attention network for knowledge graph completion[J].Neural Networks, 2022, 154: 234-245.

[22] Tan Zhaoxuan, Chen Zilong, Feng Shangbin, et al. KRACL: contrastive learning with graph context modeling for sparse knowledge graph completion[C]//Proceedings of the ACM Web Conference 2023. New York, USA: ACM, 2023: 2548-2559.

[23] Gregucci C, Nayyeri M, Hernández D, et al. Link prediction with attention applied on multiple knowledge graph embedding models[C]//Proceedings of the ACM Web Conference 2023. New York, USA: ACM, 2023: 2600-2610.

[24] Lee J, Wang Yequan, LI Jing, et al. Multimodal reasoning with multimodal knowledge graph[C]//Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics. Stroudsburg, PA, USA: ACL, 2024: 10767-10782.

[25] Jian Yue, Luo Xiangyu, Li Zhifei, et al. APKGC: noise—enhanced multi—modal knowledge graph completion with attention penalty[C]//Proceedings of the Thirty—Ninth AAAI Conference on Artificial Intelligence and Thirty—Seventh Conference on Innovative Applications of Artificial Intelligence and Fifteenth Symposium on Educational Advances in Artificial Intelligence. Palo Alto, CA, USA: AAAI Press, 2025: 15005-15013.

[26] Lu Yuhuan, Yu Weijian, Jing Xin, et al. HyperFM: fact—centric multimodal fusion for link prediction over hyper—relational knowledge graphs[C]//Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics. Stroudsburg, PA, USA: ACL, 2025: 2818-2830.

[27] Shang Bin, Zhao Yinliang, Liu Jun, et al. Mixed geometry message and trainable convolutional attention network for knowledge graph completion[C]//Proceedings of the Thirty—Eighth AAAI Conference on Artificial Intelligence and Thirty—Sixth Conference on Innovative Applications of Artificial Intelligence and Fourteenth Symposium on Educational Advances in Artificial Intelligence. Palo Alto, CA, USA: AAAI Press, 2024: 8966-8974.

[28] Li Guangyao, Sun Zequn, Hu Wei, et al. Position—aware relational transformer for knowledge graph embedding[J].IEEE Transactions on Neural Networks and Learning Systems, 2024, 35(8): 11580-11594.

[29] Liu Yuhan, Cao Zelin, Gao Xing, et al. Bridging the space gap: unifying geometry knowledge graph embedding with optimal transport[C]//Proceedings of the ACM Web Conference 2024. New York, USA: ACM, 2024: 2128-2137.

[30] Yin Hong, Zhong Jiang, Li Rongzhen, et al. High—order neighbors aware representation learning for knowledge graph completion[J].IEEE Transactions on Neural Networks and Learning Systems, 2025, 36(3): 5273-5287.

[31] Yang B, Yih S W, He X, et al. Embedding entities and relations for learning and inference in knowledge bases[EB/OL]. (2014—12—20)[2025—12—05].https://arxiv.org/abs/1412.6575.

[32] Szegedy C, Vanhoucke V, Ioffe S, et al. Rethinking the inception architecture for computer vision[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, NJ, USA: IEEE, 2016: 2818-2826.

[33] Kingma D P . Adam: A method for stochastic optimization[EB/OL]. (2014—12—22)[2025—12—05].https://arxiv.org/abs/1412.6980.

[34] Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: a simple way to prevent neural networks from overfitting[J].The Journal of Machine Learning Research, 2014, 15(1): 1929-1958.

[35] Ioffe S, Szegedy C . Batch normalization: accelerating deep network training by reducing internal covariate shift[C]//Proceedings of the 32nd International Conference on International Conference on Machine Learning. Chia Laguna Resort, Sardinia, Italy: PMLR, 2015: 448-456.

[36] Ge Xiou, Wang Yuncheng, Wang Bin, et al. Compounding geometric operations for knowledge graph completion[C]//Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics. Stroudsburg, PA, USA: ACL, 2023: 6947-6965.

基金资助

国家自然科学基金资助项目(62293553)