Experiments on chemdner

[antct]

Hi, I ran experiments on dataset chemdner and could not get the model performance reported in the paper.
I follow the paper and modify the topk to 60 and the number of seed rules to 40, can you provide more details about it? Thanks.
My final model f1 is relatively low (~50.0), this is my run script:

python train_demo.py --dataset chemdner --encoder scibert