MANNER: A Variational Memory-Augmented Model for Cross Domain Few-Shot Named Entity Recognition

This repository contains the official implementation of our paper MANNER: A Variational Memory-Augmented Model for Cross Domain Few-Shot Named Entity Recognition (ACL 2023).

Introduction

MANNER is a variational memory-augmented few-shot NER model. It uses a memory module to store information from the source domain and then retrieve relevant information from the memory to augment few-shot tasks in the target domain. Moreover, in order to effectively leverage the information from memory, MANNER uses optimal transport to retrieve and process information from the memory.

File Structure

data: contains the datasets used in the experiments.

cross_dataset: contains the code for the English Cross-Domain dataset.

chinese_cross_dataset: contains the code for the Chinese Cross-Domain dataset.

saved_models: contains the saved model checkpoints.

scripts: contains some example scripts for training and evaluation.

Requirements

torch=1.7.1
transformers=3.4.0
scikit-learn=0.24.2
pot=0.7.0
seqeval 

Datasets

For English cross-domain datasets, you can download the dataset from here. In addition to the dataset, you also need to download the entity types of all the datasets from here. You can put the datasets and the entity types in the data folder.

Training and Evaluation

We provide some examples of how to train and evaluate MANNER in the scripts folder. For example, if you want to train and evaluate the model on the ontonotes dataset in the 1-shot setting, you can simply run the command:

bash scripts/ontonotes_1shot.sh 

Please refer to the appendix of our paper for the hyperparameter settings on each dataset.

Note: I am still cleaning the code for the Chinese Cross-Domain dataset and will upload the code soon.

---

If you found the paper and the resources useful, please kindly cite our paper:

@inproceedings{fang2023manner,
  title={MANNER: A Variational Memory-Augmented Model for Cross Domain Few-Shot Named Entity Recognition},
  author={Fang, Jinyuan and Wang, Xiaobin and Meng, Zaiqiao and Xie, Pengjun and Huang, Fei and Jiang, Yong},
  booktitle={Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  pages={4261--4276},
  year={2023}
}

Contact:

If you have any questions, feel free to contact me via (fangjy6@gmail.com).