基于信息传播节点集的CTDN节点分类算法  

Node Classification Algorithm Based on Information Propagation Node Set for CTDN

作　　者：黄鑫 李赟 熊瑾煜 HUANG Xin;LI Yun;XIONG Jinyu(College of Information System Engineering,PLA Strategic Support Force Information Engineering University,Zhengzhou 450001,China;National Key Laboratory of Science and Technology on Blind Signal Processing,Chengdu 610041,China)

机构地区：[1]中国人民解放军战略支援部队信息工程大学信息系统工程学院,郑州450001 [2]盲信号处理国家级重点实验室,成都610041

出　　处：《计算机工程》2021年第6期188-196,共9页Computer Engineering

基　　金：国防科技重点实验室基金。

摘　　要：针对连续时间动态网络的节点分类问题,根据实际网络信息传播特点定义信息传播节点集,改进网络表示学习的节点序列采样策略,并设计基于信息传播节点集的连续时间动态网络节点分类算法,通过网络表示学习方法生成的节点低维向量以及OpenNE框架内的LogicRegression分类器,获得连续时间动态网络的节点分类结果。实验结果表明,与CTDNE和STWalk算法相比,该算法在实验条件相同的情况下,网络表示学习结果的二维可视化效果更优且最终的网络节点分类精度更高。The study described in this paper addresses the problem of node classification in Continuous-Time Dynamic Network(CTDN).In this work,an information propagation node set is defined according to the features of the actual network information propagation,and the node sequence sampling strategy in network representation learning is improved.Based on the defined information propagation node set,a node classification algorithm for CTDN is designed.The algorithm employs the network representation method to generate the low-dimensional node vector,and uses the LogicRegression classifier to obtain the node classification results of CTDN.Experimental results show that the proposed algorithm outperforms the existing classic algorithms such as CTDNE and STWalk under the same experimental conditions,providing better 2D visualized network representation learning results and higher network node classification accuracy.

关 键 词：信息传播节点集 连续时间动态网络 网络表示学习 节点分类 随机游走 Skip-Gram模型 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]