Hi Matthieu, thanks for the comment, it's a good quesiton to make the article more accessible.

So graph-focused means that we have a labelled graph (for example a graph with all users from twitter which are fake). We can train a model on this graph and return predictions on new graphs to check whether they are 'fake' or not.

Node-focus approach means that we want predicitons on each node. In this case, nodes have their own label and we train a model to get predictions at node-level.

Let me know if this helps :)