The framework can handle all labeled data？

[tanjia123456]

hello, I want to use my own datasets, but unfortunately all data are labeled. So, can I use this framework to handle all labeled data?

[scottjiao]

Yes it can handle all the dataset which normal GCN can handle. However, this framework fits well for few labels data, if the label information is too ample, I suggest to directly use normal GCN as alternative.

[tanjia123456]

Actually, I haven't found a code that is suitable for all marked, many are unsupervised or semi-supervised. If you know, please tell me. In short, thank you for your reply

[scottjiao]

Graph neural networks are designed for semi-supervised problems with graph structures, especially for the node classification problems. If you have a strong need to deal with all-marked dataset, maybe traditional methods such as random-walk based methods, e.g. node2vec, or matrix factorization methods, e.g. pagerank or label propagation, are better.

[tanjia123456]

Sorry, my understanding is that I can use a semi-supervised framework and then add random walks. Is this correct?

------------------ 原始邮件 ------------------ 发件人: "scottjiao/Dynamic-selftraining-GCN" <notifications@github.com>; 发送时间: 2020年9月24日(星期四) 下午4:29 收件人: "scottjiao/Dynamic-selftraining-GCN"<Dynamic-selftraining-GCN@noreply.github.com>; 抄送: "1248742467"<1248742467@qq.com>;"Author"<author@noreply.github.com>; 主题: Re: [scottjiao/Dynamic-selftraining-GCN] The framework can handle all labeled data？ (#1)

Graph neural networks are designed for semi-supervised problems with graph structures, especially for the node classification problems. If you have a strong need to deal with all-marked dataset, maybe traditional methods such as random-walk based methods, e.g. node2vec, or matrix factorization methods, e.g. pagerank or label propagation, are better.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.