train.json

[alexsimonov1]

Hey, can you please explain how you obtained head, pos, and deprel in JSON files? In the raw file of SemEval, there is no such information, but in your code, you already have these attributes, which were not created in the code itself.
As I understood, these attributes should be created based on the LAL parser, but in the DualGCN code, this parser is used only to obtain the adjacency matrix and syntree.

[sdwulxr]

您好，我想问一下，您运行出来结果了吗？
我运行出来的结果和论文中差了2%左右，而且 parser.add_argument('--losstype', default=None, type=str, help="['doubleloss', 'orthogonalloss', 'differentiatedloss']")这一句将损失类型改成doubleloss之后更低了是为什么呢？不知道您有没有遇到这个问题，期待您的回复，谢谢！