tangminji/STGN-NoisyNER

Sub-experiment of "STGN: an Implicit Regularization Method for Learning with Noisy Labels in Natural Language Processing" (EMNLP 2022) by Tingting Wu, Xiao Ding, Minji Tang, Hao Zhang, Bing Qin, Ting Liu.

STGN Experiment on NoisyNER

Sub-experiment of "STGN: an Implicit Regularization Method for Learning with Noisy Labels in Natural Language Processing" (EMNLP 2022) by Tingting Wu, Xiao Ding, Minji Tang, Hao Zhang, Bing Qin, Ting Liu.

Main experiment: github

Experiment on SST, NoisyNER and wikiHow:

• SST:

  • our code

• NoisyNER:

  • our code
  • original work

• wikiHow:

  • our code
  • original work

Data: NoisyNER

Data Source: "Analysing the Noise Model Error for Realistic Noisy Label Data" (AAAI 2021) by Hedderich, Zhu & Klakow

This experiment was adapt from "https://github.com/uds-lsv/noise-estimation/tree/master/base_model_performance/exp_ner", and you should follow the instruction of original README.

Run Code

We give an example shell:

i=1
method=base

for j in 0 1 2
do
    python tm_train_hy_nruns.py \
    --dataset NoisyNER \
    --exp_name ../nrun/ner_$method/$i/seed$j/ \
    --params_path best_params$i.json \
    --out_tmp ner_out_tmp$i.json \
    --sub_script sbatch_ner_hy_sub$i.sh \
    --output_dir ../logger/ \
    --label_set $i \
    --seed $j
done

Options:

• label_set(i): 1,2,3,4,5,6,7
• method: base, STGN

You should run the tm_train_hy_nruns.py with proper params written in json files.