Closed tanikina closed 3 months ago
Oh, that looks very cool!
One question: In what order did you arrange the nodes? Do the node ids imply any order, e.g. in means of how the (original) text was created? Because I've seen this in your code, but the graph is a bit hard to read and I was wondering if it is the original data or the order how it is visualized. Can you elaborate how you did it?
Thank you! Actually, in this particular example (nodeset 18299) even if we comment the sorting out we will still get the same output graph :)
Empirically, I found that sorting by the node id makes graphs look a bit better (they have shorter edges if the nodes are closer to each other) and I think that the overall complexity stems from the original data + GraphViz magic/constraints. The ids in the original data seem to follow the dialogue turn order but I also found that some nodes were not connected to anything at all (i.e., they were w/o any edges!), e.g., node 541666 in the nodeset 18299.
PS: According to their annotation guidelines: I-nodes are propositions (they express information) and L-nodes are locutions. I took the node type annotations directly from the data.
One more comment: Can we have the I-nodes on the left and the L-nodes on the right so that it follows the IAT diagram schema? I think it was introduced in Budzynska, Katarzyna et al. “Towards Argument Mining from Dialogue.” Comma (2014).
Agreed, I updated the order and also fixed some missing edges between the TA and S nodes in the previous version. The current version looks as follows (I left only time in the timestamps since the date is the same for the whole graph):
I really hope that I did not forget anything this time 🙃
This adds nodeset (map) visualizations based on the JSON files of the Shared Task using Graphviz library.
Example for nodeset 18299:
This should generate the following visualization:
Please let me know if this is useful and whether I should change something.