基于几何邻居的半监督节点分类

Semi-supervised node classification based geometric neighbor

目前基于网络结构的节点分类方法只注重局部网络连接关系。为了能获取更广泛的网络信息,提出一种基于邻居节点结构信息的半监督节点分类算法CBGN。首先,在网络中加入惩罚因子来改进随机游走策略以获取节点的不定长游走序列,这些节点序列被当做句子输入到word2vec模型中,从而将网络结构的潜在信息转换成向量作为节点的特征表示;其次,改进支持向量机算法,结合梯度下降法和坐标下降法来优化参数空间,以对未标记节点进行更准确的分类;最后,在四个标准数据集上与目前较先进的几种方法进行了对比实验。结果表明,CBGN算法提高了分类精度,相比之前已有的方法具有更好的分类效果。

The existent node classification methods based on network structure only pay attention to local network connection relationship. For obtaining wider network information,this paper developed a semi-supervised node classification algorithm( CBGN) based on geometric neighbor structure information. This algorithm improved random walk strategy with penalty factor in network to obtain arbitrary length node sequence for each node. It input these node sequences into the word2 vec model for transforming the potential information into node vectors. CBGN combined gradient descent method and the coordinate descent method to optimize the SVM classification model. This method compared with current methods on four standard datasets. The results verify that the proposed algorithm improves the classification accuracy and has better classification effect.