search

jasperzhong / read-papers-and-code

My paper/code reading notes in Chinese
43 stars 3 forks source link

SIAM '22 | Heterogeneous Temporal Graph Neural Network #335

Closed jasperzhong closed 1 year ago

jasperzhong commented 1 year ago

https://epubs.siam.org/doi/pdf/10.1137/1.9781611977172.74

jasperzhong commented 1 year ago

DTDG-based的方法，和DySAT很像.

image

这图里intra-relation和inter-relation不就是HAN #328 嘛？相当于是HAN作用于每个graph snapshot生成T个node embeddings，这些ndoe embeddings加上一个time encoding，做一个attention，得到最终的node embedding. 非常straightforward.

和GATNE #341 和DyHART #344 差不多...先分relation作aggregation，然后aggregate不同relation，最后aggregate各个timestamp的...真的大同小异.

最后实验是在OGBN-MAG（分成10个graph snapshots）和COVID-19（分成304个graph snapshots），相比static models和之前的dynamic models取得了SOTA的效果.

image

jasperzhong commented 1 year ago

A notable detail: for OGBN-MAG dataset, they use metapath2vec #330 to generate 128-dim node embedding as input features for nodes without features.