Building text graph for GNN

I was wondering if you could help me on building text graph for text classification using GNN with dgl.

The key part of using GNN for text classification is that you need to build the graph by yourself instead of taking it from the dataset. Different from the work on citation graphs or molecular graphs, there is no golden standard for constructing a text graph. So you need to build a text graph first, and then apply a GNN like the GCN or GAT as usual. For example, you can try method in [1809.05679] Graph Convolutional Networks for Text Classification to build your graph. However, you need to trade off between the performance (encourages you to have a dense graph) and the scalability (you need a sparse graph).

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.