摘要
鉴于现实中存在权限设置,获取数据成本过高等因素,大多数收集到的网络仅为完整网络的局部,检测缺失的节点是处理网络不完整问题的有效途径之一。网络缺失节点的检测是一个复杂的问题,因为其不仅需要判断网络是否缺失了节点,还需要分析与缺失节点相关的边。于是提出了基于度量学习的网络缺失节点检测模型(MND-M),它利用图卷积神经网络分别学习两个样本,学习的结果用于预测每组样本内是否缺失了节点,并利用度量学习方法比较一对样本嵌入的相似度,判断它们缺失的节点是否为同一个。模型在完整网络中构造了样本数据及标签来训练模型,帮助其获得缺失节点检测的能力。在4个数据集上进行了实验,结果表明MND-M能有效检测出网络中节点,获得较高的准确率和F1值。
Due to the phenomenon of permission setting and high cost of data acquisition,most of the collected networks are only partial parts of the complete network.Detecting the missing nodes is one of the effective ways to deal with the problem of incomplete network.The detection of missing nodes in networks is a complex problem,because it not only needs to judge whether the network is missing nodes,but also needs to analyze the edges related to missing nodes.Therefore,a model for detecting missing nodes in network based on metric learning(MND-M)is proposed,which uses graph convolution neural networks to learn two samples separately.The learning results are used to predict whether a node is missing within each of samples.The metric learning method is then used to compare the similarity of the pair of sample embeddings.The similarity is used to determine whether the missing nodes of a pair of samples are the same.Sample data and labels are constructed in the full network to train the model and allow the model to gain the capability of missing node detection.Experiments are conducted on 5 datasets.The experimental results show that MND-M can effectively detect nodes in the network,obtaining high accuracy and F1 scores.
作者
刘臣
邵颖
周立欣
LIU Chen;SHAO Ying;ZHOU Lixin(School of Business,University of Shanghai for Science and Technology,Shanghai 200093)
出处
《计算机与数字工程》
2022年第11期2526-2532,共7页
Computer & Digital Engineering
基金
上海市哲学社会科学规划课题(编号:2021BTQ003)资助。
关键词
度量学习
图卷积神经网络
缺失节点
网络数据
深度学习
metric learning
graph convolution neural networks
missing node
network data
deep learning
