Dear Thomas, i appreciate your insights on the GAE. GAE use GCN get the embdding of the graph then use it do a link prediction task. Compared with the initial matrix, the reconstructed matrix becomes dense.
In my task, the initial matrix is dense, by take node features as input, the expect reconstruct matrix become saprse, e.g. many nodes will lost their links. However, I did an experiment and found that the effect of the current model is very poor, the reconstructed adjacency matrix becomes particularly dense. So i ask your advice on how to deal with it. Look forward to your favourable reply.
Dear Thomas, i appreciate your insights on the GAE. GAE use GCN get the embdding of the graph then use it do a link prediction task. Compared with the initial matrix, the reconstructed matrix becomes dense.
In my task, the initial matrix is dense, by take node features as input, the expect reconstruct matrix become saprse, e.g. many nodes will lost their links. However, I did an experiment and found that the effect of the current model is very poor, the reconstructed adjacency matrix becomes particularly dense. So i ask your advice on how to deal with it. Look forward to your favourable reply.