《计算机应用研究》|Application Research of Computers

保持聚类结构的属性网络表示学习

Clustering-preserving representation learning on attributed network

免费全文下载 (已被下载 次)  
获取PDF全文
作者 张静,柴变芳,张璞,李文斌
机构 河北地质大学 a.信息工程学院;b.教务处,石家庄 050031
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2020)06-008-1647-05
DOI 10.19734/j.issn.1001-3695.2018.12.0879
摘要 在线社交平台产生大量可建模为属性网络的数据,SNE(social network embedding)表示学习模型可学到属性网络的潜在低维表示,为进一步的实际应用提供有效特征。但是SNE未考虑保持网络的潜在聚类结构,导致学到的特征对聚类效果不佳。对此提出了一种保持聚类结构的属性网络表示学习模型(attributed network embedding with self cluster,ANESC),其使用前馈神经网络建模,以属性网络节点的one-hot表示和属性信息作为输入,经过多隐层学习节点的低维表示,使其在输出层保持节点的邻居拓扑结构和潜在聚类结构。在五个真实属性网络上的实验结果表明,相比SNE,ANESC学到的表示在聚类任务上NMI值提高5%~11%,在分类任务上准确率提高0.3%~7%。
关键词 网络表示学习; 属性网络; 前馈神经网络
基金项目 国家自然科学基金资助项目(61503260)
河北省研究生创新资助项目(CXZZSS2018118)
本文URL http://www.arocmag.com/article/01-2020-06-008.html
英文标题 Clustering-preserving representation learning on attributed network
作者英文名 Zhang Jing, Chai Bianfang, Zhang Pu, Li Wenbin
机构英文名 a.School of Information Engineering,b.Academic Affairs Office,Hebei Geo University,Shijiazhuang 050031,China
英文摘要 The online social platform generates a lot of data that can be modeled as attributed network. The representation learning model of SNE(social network embedding) can learn the potential low-dimensional representation of the attributed network, which provides effective features for further practical applications. However, SNE does not consider preserving the potential clustering structure of the network, which results in bad clustering effect. In order to solve these problems, this paper proposed an attributed network representation learning model(attributed network embedding with self cluster, ANESC) that preserved the clustering structure, which used feedforward neural network to model. ANESC took one-hot representation and attributed information of attributed network nodes as input, with multi-hidden layer learning node′s low-dimensional representation, it preserved the node's neighbor topology and potential clustering structure at the output layer. The empirical results of five real attributed networks show that compared with the presentation that SNE learns, the NMI value of presentation that ANESC learns in clustering task increases by 5%~11%, and the accuracy of classification increases by 0.3%~7%.
英文关键词 network representation learning; attributed network; feedforward neural network
参考文献 查看稿件参考文献
 
收稿日期 2018/12/11
修回日期 2019/1/25
页码 1647-1651
中图分类号 TP181
文献标志码 A