Closed pangsg closed 3 years ago
Thank you so much.
------------------ 原始邮件 ------------------ 发件人: "TianBian95"<notifications@github.com>; 发送时间: 2020年10月10日(星期六) 下午5:24 收件人: "TianBian95/BiGCN"<BiGCN@noreply.github.com>; 抄送: "Yaphets"<594417019@qq.com>; "Author"<author@noreply.github.com>; 主题: Re: [TianBian95/BiGCN] GCN能用于有向图吗 (#1)
For the usage of GCN on a directed graph, actually, GCN's author added the undirected assumption to the GCN so that the GCN model can be explained in terms of spectral analysis. However, it does not mean that the GCN model can only deal with undirected graphs. In fact, GCN could also be viewed as a special message-passing model. From the perspective of the message-passing model, either an undirected graph or a directed graph can be an input of GCN, as described in Remark 2 on page 6 of [1]. There have been a few papers explaining the GCN model from the perspective of the message-passing framework although the theoretical analysis of GCN on a directed graph is still under research. For example, in [2] mentioned by GCN's author at tkipf/gcn#91, the GCN model is understood as a special case of a simple differentiable message-passing framework as equation (1).
[1] Scarselli, F. et. al. 2008. The graph neural network model. IEEE Transactions on Neural Networks, 20(1), 61-80. [2]Schlichtkrull, M. et. al. 2018. Modeling relational data with graph convolutional networks. ESWC (pp. 593-607). Springer, Cham.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.
For the usage of GCN on a directed graph, actually, GCN's author added the undirected assumption to the GCN so that the GCN model can be explained in terms of spectral analysis. However, it does not mean that the GCN model can only deal with undirected graphs. In fact, GCN could also be viewed as a special message-passing model. From the perspective of the message-passing model, either an undirected graph or a directed graph can be an input of GCN, as described in Remark 2 on page 6 of [1]. There have been a few papers explaining the GCN model from the perspective of the message-passing framework although the theoretical analysis of GCN on a directed graph is still under research. For example, in [2] mentioned by GCN's author at https://github.com/tkipf/gcn/issues/91, the GCN model is understood as a special case of a simple differentiable message-passing framework as equation (1).
[1] Scarselli, F. et. al. 2008. The graph neural network model. IEEE Transactions on Neural Networks, 20(1), 61-80. [2]Schlichtkrull, M. et. al. 2018. Modeling relational data with graph convolutional networks. ESWC (pp. 593-607). Springer, Cham.