基于二阶图自编码器的复杂网络分析

Analysis of Complex Network Based on Second-order Graph Autoencoder

为了充分利用复杂网络中蕴含的信息,增强图自编码器模型的表征能力,提出一种基于二阶图卷积网络的自编码器模型SeGCN-AE。先使用二阶图卷积网络提取实体属性和关系信息,生成低维特征表示;然后使用内积解码器重构复杂网络链接关系矩阵,并通过重构损失对模型进行优化。在两个基准复杂网络数据集实验中,SeGCN-AE的性能始终优于当前较为先进的基线模型,表明二阶关系的引入能够增强模型的表征能力,提升复杂网络分析任务的表现。

In order to make full use of the information contained in complex networks and enhance the representation ability of graph autoencoder models,we propose an autoencoder model SeGCN-AE based on Second-order Graph Convolutional Networks(SeGCN).First,SeGCN is used to extract entity attributes and relationship information,and generate low-dimensional feature representations.Then,the inner product decoder is used to reconstruct the complex network link relationship matrix,and the model is optimized by reconstruction loss.On the two baseline complex network dataset experiments,the performance of SeGCN-AE is always better than current advanced baseline model,indicating that the introduction of second-order relationships can enhance representation ability of the model and improve the performance of complex network analysis tasks.