ESD

Code For NAACL2022 Paper "An Enhanced Span-based Decomposition Method for Few-Shot Sequence Labeling"

🔥 Introduction

Few-Shot Sequence Labeling (FSSL) is a canonical paradigm for the tagging models, e.g., named entity recognition and slot filling, to generalize on an emerging, resource-scarce domain. Recently, the metric-based meta-learning framework has been recognized as a promising approach for FSSL. However, most prior works assign a label to each token based on the token-level similarities, which ignores the integrality of named entities or slots. To this end, in this paper, we propose ESD, an Enhanced Span-based Decomposition method for FSSL. ESD formulates FSSL as a span-level matching problem between test query and supporting instances.Specifically, ESD decomposes the span matching problem into a series of span-level procedures, mainly including enhanced span representation, class prototype aggregation and span conflicts resolution. Extensive experiments show that ESD achieves the new state-of-the-art results on two popular FSSL benchmarks, FewNERD and SNIPS, and is proven to be more robust in the nested and noisy tagging scenarios.

🚀 How to use our code?

💾 Environment

pip install -r requirements.txt

We run our code on a single NVDIA A40 GPU, and we use torch1.8.0+cu111:

pip install torch==1.8.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html

📚 Dataset

mkdir data
cd data

FewNERD

We use the latest sampled dataset from FewNERD, which corresponds to the results of the 6-th version FewNERD paper. The new sampled data fixs the data sampling bug, see issue.

wget -O data.zip https://cloud.tsinghua.edu.cn/f/0e38bd108d7b49808cc4/?dl=1
unzip data.zip
mv episode-data/* ./
rm -rf episide-data

SNIPS

We use the sampled data from SNIPS-FewShot.

wget https://atmahou.github.io/attachments/ACL2020data.zip
unzip ACL2020data.zip
mv ACL2020data/* ./
rm -rf ACL2020data
cd ..

🏋🏻‍♂️ Train and Evaluation

mkdir checkpoint
mkdir results

Make sure your have the data file structure as follows:

├── bash
│   ├── fewnerd
│   └── snips
├── checkpoint
├── data
│   ├── inter
│   ├── intra
│   ├── xval_snips
│   └── xval_snips_shot_5
├── model
│   ├── ESD.py
│   └── utils.py
├── README.md
├── requirements.txt
├── results
├── train_demo.py
└── util
    ├── data_loader.py
    ├── framework.py
    └── utils.py

FewNERD

bash bash/fewnerd/run_mode.sh [gpu_id] [mode] [N] [K]
    - mode: intra/inter
    - N, K: 5 1, 5 5, 10 1
    e.g., bash bash/fewnerd/run_mode.sh 0 inter 5 1
bash bash/fewnerd/10wat_5shot_mode.sh [gpu_id] [mode]
    - mode: intra/inter
    e.g., bash/fewnerd/10wat_5shot_mode.sh 0 inter

SNIPS

bash bash/snips/1-shot/1_shot_mode_1.sh [gpu_id]
...
bash bash/snips/1-shot/1_shot_mode_7.sh [gpu_id]

bash bash/snips/5-shot/5_shot_mode_1.sh [gpu_id]
...
bash bash/snips/5-shot/5_shot_mode_7.sh [gpu_id]

🌝 Citation

If you use our code, please cite our paper:

@inproceedings{wang-etal-2022-enhanced,
    title = "An Enhanced Span-based Decomposition Method for Few-Shot Sequence Labeling",
    author = "Wang, Peiyi  and
      Xu, Runxin  and
      Liu, Tianyu  and
      Zhou, Qingyu  and
      Cao, Yunbo  and
      Chang, Baobao  and
      Sui, Zhifang",
    booktitle = "Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
    month = jul,
    year = "2022",
    address = "Seattle, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.naacl-main.369",
    doi = "10.18653/v1/2022.naacl-main.369",
    pages = "5012--5024",
    abstract = "Few-Shot Sequence Labeling (FSSL) is a canonical paradigm for the tagging models, e.g., named entity recognition and slot filling, to generalize on an emerging, resource-scarce domain. Recently, the metric-based meta-learning framework has been recognized as a promising approach for FSSL. However, most prior works assign a label to each token based on the token-level similarities, which ignores the integrality of named entities or slots. To this end, in this paper, we propose ESD, an Enhanced Span-based Decomposition method for FSSL. ESD formulates FSSL as a span-level matching problem between test query and supporting instances. Specifically, ESD decomposes the span matching problem into a series of span-level procedures, mainly including enhanced span representation, class prototype aggregation and span conflicts resolution. Extensive experiments show that ESD achieves the new state-of-the-art results on two popular FSSL benchmarks, FewNERD and SNIPS, and is proven to be more robust in the noisy and nested tagging scenarios.",
}