search

yao8839836 / text_gcn

Graph Convolutional Networks for Text Classification. AAAI 2019
1.34k stars 435 forks source link

Why to multiply A_hat with X or L_i #9

Open ghost opened 5 years ago

ghost commented 5 years ago

Hello, I have a theoretical problem, in the method part of your paper, that why to multiply A(normalized) with X or L_i? What's the meaning of this multiplication?

yao8839836 commented 5 years ago

Hi, @dagne-ged

Thanks for your question.

This is taken from the original GCN paper [1]. Please see the derivation in equation 3, 4, 5, 6, 7, 8.

According to [2]. The multiplication means generating a new feature matrix Y from X or L_i by applying the graph convolution:

Y = A^{hat}X

where A^{hat} is the convolution matrix. After generating the new feature matrix Y for all nodes, GCN becomes a simple fully-connected network. Please see equation 4, 7, 8 in [2].

[1] Kipf, T. N., and Welling, M. 2017. Semi-supervised classification with graph convolutional networks. In ICLR [2] Li, Q.; Han, Z.; and Wu, X. 2018. Deeper insights into graph convolutional networks for semi-supervised learning. In AAAI.

ghost commented 5 years ago

Oh oh, I get it~ Thanks a lot for your kind suggestion:）