This repository contains the implementation of our paper "PeerDA: Data Augmentation via Modeling Peer Relation for Span Identification Tasks" (ACL 2023).

• python==3.7.3
• pytorch==1.2.0 (also tested on pytorch 1.3.0)
• transformers4.1.1
• numpy==1.16.4
• tensorboardX==1.9
• tqdm==4.32.1

• roberta-base/large for NER, CCE, and SBPD
• bert-large-uncased for ABSA

• Please refer to ./$task/Data/$dataset (e.g. ./NER/Data/wnut) for the data files of each task.
• For large datasets like OntoNotes5 and News20 SBPD, please download from official websites 👉 [OntoNotes5][News20 SBPD].
• For CCE data
  1. Download the data directly from CUAD
  2. Rename the train/test splits and Place the files in CCE/Data/*

     mv train_separate_questions.json CCE/Data/mrc-cce.test
     mv test.json CCE/Data/mrc-cce.test
     

• Reproduce the results on WNUT17 (NER) dataset with RoBERTa-base:
  1. place data files in the directory ./NER/Data/wnut (we have already put the WNUT17 data in this directory. But for other large datasets, please download from our provided and put them in the corresponding data folder).
  2. Enter the task folder

     cd NER
     

  3. Train the WNUT model:

     python NER/train-NER.py \
       --output_dir ./saved_models/DA-roberta-base-wnut-1e-5 \
       --model_type roberta \
       --model_name_or_path roberta-base --cache_dir ../cache \
       --data_path ./Data/wnut \
       --do_train --do_eval --do_lower_case \
       --learning_rate 1e-5 \
       --num_train_epochs 30 \
       --per_gpu_eval_batch_size=64  \
       --per_gpu_train_batch_size=32 \
       --max_seq_length 160 --max_query_length 32\
       --save_steps 0 --logging_steps 1000\
       --fp16 --expand_rate 1 --gradient_accumulation_steps 1\
       --overwrite_output_dir --evaluate_during_training --DA # --sizeonly
     

     where --DA enables PeerDA variants:
     • PeerDA-Both: Please use the defaulted code.
     • PeerDA-Categ: Please use --expand_rate 0.
     • PeerDA-Size: Please use --expand_rate 1 and enable --sizeonly.

1. place data files in the directory ./$task/Data/$dataset if not done yet.
2. Train the model with the following scripts (Note that there is a data caching process (only once) for CCE tasks to split the contracts into training examples, which takes some time before training the model):

   cd $task
   bash $script
   

TODO

If the code is used in your research, please star our repo and cite our paper as follows:

@inproceedings{xu2022peerda,
    title = "PeerDA: Data Augmentation via Modeling Peer Relation for Span Identification Tasks",
    author = "Xu, Weiwen  and
      Li, Xin  and
      Deng, Yang  and
      Lam, Wai  and
      Bing, Lidong",
    booktitle = "The 61th Annual Meeting of the Association for Computational Linguistics.",
    year = "2023",
}