宋国杰 - 子图结构特征提取驱动的图神经网络模型.pdf

1、Graph Neural Networks with Graph Neural Networks with Substructure PatternsSubstructure Patterns GuojieGuojie SONG (SONG (宋国杰) ) School of EECS, Peking School of EECS, Peking UniversityUniversity 第九届全国社会媒体处理大会（SMP2020）中国-杭州 2 Outline Graph Neural Networks Why Care about Substructure Pattern GNN with

2、 Substructure Pattern GraLSP Model Yilun Jin, Guojie Song*, Chuan Shi. GraLSP: Graph Neural Networks with Local Structural Patterns. In Proc. of AAAI2020. GraphSTONE Model Qingqing Long, Yilun Jin, Guojie Song*, Yi Li, Wei Lin. Graph Structural- topic Neural Network. In Proc. of KDD2020. About Futur

3、e Works Network Representation Learning 3 G = ( V, E ) Vector Space Easy to parallel Apply classical ML methods generate embed What are the relationships between shallow model and deep model? Shallow Model vs Deep Model 4 Shallow ModelDeep Model Goal: : Need to define!Need to define! Shallow Model unsupervised 5 Deep Model: GNNs Graph Convolution Networks，GCN Aggregating feature information form