babyyang2049/coin_dialogre

Source codes and data for our IJCAI 2021 paper "Consistent Inference for Dialogue Relation Extraction".

CoIn_DialogRE

0. Package Description

├─ data/: raw data and preprocessed data
    ├─ train.json
    ├─ dev.json
    ├─ test.json
    ├─ entity_type_id.json
    ├─ speaker_vocab_id.json
    ├─ vocab.txt: bert vocab file, we add the new-introduced special tokens
├─ logs/: save the log files
├─ model/: save the optimal model file and prediction results
├─ src/: source codes
    ├─ attention.py 
    ├─ data_utils.py: utils for processing data
    ├─ dataset.py
    ├─ embeddings.py: generate entity type/ utterance embedding
    ├─ model.py
    ├─ main.py: main file to run the model
├─ readme.md

1. Environments

We conducted experiments on a sever with two GeForce GTX 1080Ti GPU.

• python (3.6.5)
• cuda (11.0)
• CentOS Linux release 7.8.2003 (Core)

2. Dependencies

• torch (1.2.0)
• transformers (2.0.0)
• pytorch-transformers (1.2.0)
• numpy (1.19.2)

3. Preparation

3.1 Download the pre-trained language models.

• Download the bert-base-uncase model.

3.2 Add the special token id into the vocab.txt

• Inspired by the resource paper, we add the newly-introduced special tokens to indicate the speakers. (Replacing [unused1]..[unsued10] with speaker1..speaker10).
• You can replace the original vocab.txt with our file (in './data/vocab.txt')

4. Training

If you want to reproduce our results, please follow our hyper-parameter settings and run the code with the following command.

CUDA_VISIBLE_DEVICES=0,1 nohup python -m torch.distributed.launch --nproc_per_node=2 main.py --bert_path {your_bert_path}

5. Evaluating

You also can evaluate our model without training. Please download the released model. model

python evaluate.py --bert_path {your_bert_path} --optimal_model_path {released_model_path}

Citation

Thank you for your interests in our paper, if you have any problem, please feel free to contact me. ([email protected])

@inproceedings{DBLP:conf/ijcai/LongNL21,
  author    = {Xinwei Long and
               Shuzi Niu and
               Yucheng Li},
  title     = {Consistent Inference for Dialogue Relation Extraction},
  booktitle = {Proceedings of the Thirtieth International Joint Conference on Artificial
               Intelligence, {IJCAI} 2021, Virtual Event / Montreal, Canada, 19-27
               August 2021},
  pages     = {3885--3891},
  year      = {2021},
  url       = {https://doi.org/10.24963/ijcai.2021/535}
}