search

Nealcly / templateNER

Source code for template-based NER
208 stars 39 forks source link

Cross-domain Few-Shot NER Result #17

Open lvminghan1997 opened 2 years ago

lvminghan1997 commented 2 years ago

in no source-domain data，MIT Movie 10 shot ,the result in paper is 37.3,why i get 51.06 , can you tell me some detail about that?

lplping commented 2 years ago

I have the same doubt, i get 0.496 in MIT Movie 10 shot

zhanghaok commented 2 years ago

I have the same doubt, i get 0.496 in MIT Movie 10 shot

您好！我现在已经在CoNLL03上复现了你的结果，也生成了对应的模型，请问我如何在MIT Movie少样本数据集上进行微调呢，直接在train.py中修改数据集的路径，然后运行就可以了吗?

zhanghaok commented 2 years ago

您好！我现在已经在CoNLL03上复现了你的结果，也生成了对应的模型，请问我如何在MIT Movie少样本数据集上进行微调呢，直接在train.py中修改数据集的路径，然后运行就可以了吗?

Nealcly commented 2 years ago

One possible reason is the sample details. Could you sample multiple times and calculate the average performance?

bwl666 commented 1 year ago

in no source-domain data，MIT Movie 10 shot ,the result in paper is 37.3,why i get 51.06 , can you tell me some detail about that?

Hello, I would like to ask if dev .txt is not needed in cross-domain experiments such as mit-movie. If not, will this overfit?