search

universal-ie / UIE

Unified Structure Generation for Universal Information Extraction
900 stars 99 forks source link

The SSI difference between train and predict #14

Closed xinjicong closed 2 years ago

xinjicong commented 2 years ago

想问一下，在训练时候text前面加上的SSI是每条样本中的SSI，predict的时候是所有SSI的拼接？ 如果SSI数量过多，预测的时候需要进行SSI划分，分开预测？

luyaojie commented 2 years ago

我们的实验在训练和predict都保留了所有的SSI，请参考论文的表格11。 We use the same SSI during training and inference, details see in Table11 in the paper.

如果类别过多，可以将SSI进行划分，在训练和测试时都需要，保持两个阶段一致就可以。

luyaojie commented 2 years ago

如果有问题，欢迎再次打开Issue，或者给我发邮件 If you have any further questions, please open the Issue again or email me