基于半监督聚类的网络嵌入方法

GEMSEC(graph embedding with self clustering)在计算节点特征的同时学习节点聚类,通过强制将节点进行聚类来揭露网络中的社区结构,但未考虑类别标签信息,导致学到的节点嵌入缺乏区分性。针对这一问题,提出了一种基于半监督聚类的网络嵌入方法(NESSC),将随机游走序列和少量节点类别标签作为输入,在计算节点特征和学习节点k-means聚类的过程中,利用类别标签信息指导聚类过程,同时重构已知节点类别标签信息,学习具有区分性的节点表示。在6个真实网络上进行节点聚类和节点分类评测实验,实验结果显示,NESSC方法明显优于无监督网络嵌入方法DeepWalk和GEMSEC,可以通过加入节点的标签信息来提高网络嵌入的效果。因此,通过网络节点的嵌入,可以高效地提取网络的有用信息,对于相关网络嵌入研究具有一定的参考价值。

Network embedding algorithm based on semi-supervised clustering

GEMSEC(craph embedding with self clustering) learns a clustering of the nodes simultaneously with computing their features, and reveals the natural community structure in the network by enforcing nodes to be clustered. However, it fails to consider label information, which leads to the lack of discrimination of learned node embedding. To solve this problem, a network embedding method based on semi-supervised clustering(NESSC) is proposed, which takes the random walk sequence and a small amount labels of nodes as input, and the label information is used to guide the clustering process in the process of learning node features and node clustering learning, and the known node label information is reconstructed to learn the discriminative node representation. Finally, an experimental evaluation of node classification is performed, using six real-world datasets. The results demonstrate that NESSC is obviously superior to unsupervised network embedding methods DeepWalk and GEMSEC, which indicates that the effect of network embedding can be improved by adding the label information of nodes. This method can extract useful internet information effectively by embedding of network nodes, which provides reference for the study of network embedding methods.