Role-Based Network Representation Learning Method

doi:10.19678/j.issn.1000-3428.0057421

Abstract

Abstract: Network representation learning is widely used to obtain the characteristics and semantics of network nodes. The existing network representation learning methods mainly study the adjacency matrix or the power of the adjacency matrix,making a node in the vector space have similar nodes in the local area approximate to it in the network,but they usually ignore the structural equivalence of the global area.According to role information,this paper proposes a model called Role-Based Matrix Factorization(Role-MF) to obtain node representation.Role-MF integrates role information into a random walk method,uses role information to design a clear target matrix with local information in consideration, and obtains node representation through singular value decomposition.The experimental results show that compared with DWMF,DeepWalk and other existing models,Role-MF can retain structural equivalence,and achieves a higher F1 score and AUC in node classification and link prediction tasks when the training ratio is 10% and 90%.

Key words: role information, network representation learning, structural equivalence, matrix factorization, random walk

摘要： 网络表征学习技术被广泛应用于获取网络中节点的特征及其语义。已有网络表征学习方法主要研究邻接矩阵或邻接矩阵的幂，使得向量空间中一个节点的相似节点存在于网络中与它相近的局部区域，而未考虑全局区域的结构等价性。根据角色信息，提出基于角色的矩阵分解（Role-MF）模型来获取节点表示。Role-MF模型将角色信息融合在随机游走方法中，在考虑局部信息的同时利用角色信息设计明确的目标矩阵，并通过奇异值分解得到节点表征。实验结果表明，与现有的DWMF、DeepWalk等模型相比，Role-MF模型可以保留结构等价性，当训练比例为10%和90%时，F1值和AUC等各项指标在节点分类和链路预测中都取得了更好的效果。

关键词: 角色信息, 网络表征学习, 结构等价, 矩阵分解, 随机游走

CLC Number:

XU You, WANG Xiaoping, XIONG Yun. Role-Based Network Representation Learning Method[J]. Computer Engineering, 2021, 47(5): 52-57.

徐攸, 王晓萍, 熊贇. 基于角色的网络表征学习方法[J]. 计算机工程, 2021, 47(5): 52-57.

/ / Recommend / Download Citations

URL: http://www.ecice06.com/EN/10.19678/j.issn.1000-3428.0057421

http://www.ecice06.com/EN/Y2021/V47/I5/52

References

[1] LESKOVEC J,GROVER A.Node2Vec:scalable feature learning for networks[EB/OL].[2020-01-10].https://arxiv.org/abs/1607.00653.
[2] PEROZZI B,AL-RFOU R,SKIENA S.DeepWalk:online learning of social representations[EB/OL].[2020-01-10].https://arxiv.org/abs/1403.6652.
[3] TANG Jian,QU Meng,WANG Mingzhe.LINE:large-scale information network embedding[C]//Proceedings of International Conference on World Wide Web.Swiss Confederation,Geneva:[s.n.],2015:215-226.
[4] JENSEN D,NEVILLE J.Iterative classification in relational data[C]//Proceedings of 2000 AAAI Workshop on Learning Statistical.Austin,USA:AAAI Press,2000:466-478.
[5] WANG Xiao,CUI Peng,WANG Jing.Community preserving network embedding[C]//Proceedings of the 31st AAAI Conference on Artificial Intelligence.San Francisco,USA:AAAI Press,2017:243-259.
[6] NIE Feiping,ZHU Wei,LI Xuelong.Unsupervised large graph embedding[C]//Proceedings of the 31st AAAI Conference on Artificial Intelligence.San Francisco,USA:AAAI Press,2017:2422-2428.
[7] AKOGLU L,TONG H H,KOUTRA D.Graph based anomaly detection and description:a survey[J].Data Mining and Knowledge Discovery,2015,29(3):626-688.
[8] ZAKI M,HASAN M.A survey of link prediction in social networks[M].Boston,USA:[s.n.],2011.
[9] HEIMANN M,SHEN H M,SAFAVI T,et al.REGAL:representation learning-based graph alignment[EB/OL].[2020-01-10].https://arxiv.org/abs/1802.06257.
[10] LEVY O,GOLDBERG Y.Neural word embedding as implicit matrix factorization[C]//Proceedings of NIPS'14.Cambridge,USA:MIT Press,2014:2177-2185.
[11] RIBEIRO L,SAVERESE P,FIGUEIREDO D.Struc2Vec:learning node representations from structural identity[C]//Proceedings of KDD'17.New York,USA:ACM Press,2017:256-268.
[12] CHENG Cheng,LIU Zhiyuan,ZHAO Deli.Network representation learning with rich text information[C]//Proceedings of IJCAI'15.Buenos Aires,Argentina:[s.n.],2015:265-278.
[13] QIU Jiezhong,DONG Yuxiao,MA Hao.Network embedding as matrix factorization:unifying deep walk,LINE,PTE,and node2vec[C]//Proceedings of WSDM'18.New York,USA:ACM Press,2018:3211-3226.
[14] CAO Shaosheng,LU Wei,XU Qiongkai.Grarep:learning graph representations with global structural information[C]//Proceedings of CIKM'15.New York,USA:ACM Press,2015:306-318.
[15] CICHOCKI A,ZDUNEK R,PHAN A.Nonnegative matrix and tensor factorizations:applications to exploratory multi-way data analysis and blind source separation[M].[S.1.]:John Wiley & Sons,Ltd.,2009.
[16] HENDERSON K,GALLAPHER B,ELIASSI R.RolX:structural role extraction mining in large graphs[C]//Proceedings of KDD'12.New York,USA:ACM Press,2012:156-168.
[17] STEPHEN P,MARTIN G.Two algorithms for computing regular equivalence[J].Social Networks,1993,15(4):361-376.
[18] STANLEY W,KATHERINE F.Social network analysis:methods and applications[M].[S.1.]:Cambridge University Press,2014.
[19] LI Tao,PENG Wei.On the equivalence between nonnegative matrix factorization and probabilistic latent semantic indexing[J].Computational Statistics & Data Analysis,2011,52(8):3913-3927.
[20] KOREN Y,BELL Y,VOLINSKY C.Matrix factorization techniques for recommender systems[J].Computer,2009,42(8):30-37.
[21] OU Mingdong,CUI Peng,PEI Jian.Asymmetric transitivity preserving graph embedding[C]//Proceedings of KDD'16.New York,USA:ACM Press,2016:468-475.
[22] ZHANG Ziwei,CUI Peng,WANG Xiao.Arbitrary-order proximity preserved network embedding[C]//Proceedings of KDD'16.New York,USA:ACM Press,2018:687-696.
[23] YANG Cheng,SUN Maosheng,LIU Zhiyuan.Fast network embedding enhancement via high order proximity approximation[C]//Proceedings IJCAI'17.Melbourne,Australia:[s.n.],2017:367-378.
[24] DONNATC,MARINKA Z,DAVID H.Learning structural node embeddings via diffusion wavelets[C]//Proceedings of KDD'18.New York,USA:ACM Press,2018:269-278.

Please choose a citation manager

Content to export