Sagri-nested-ner

Domain Adaptation in Nested Named Entity Recognition From Scentific Artilces in Agriculture [SoICT2023]

Baseline

This repo is based on Triaffine-nested-ner

Environment

All codes are tested under Python 3.7, PyTorch 1.7.0 and Transformers 4.6.1. Need to install opt_einsum for einsum calculations. At least 16GB GPU are needed for training.

Make sure you have installed all required packages by running

  apt-get install build-essential -y
  pip install -r requirements.txt

Dataset

We only put 100 samples for train/dev/test. Please put datas under data/dataset_name, you can also refer word_embed.generate_vocab_embed for data paths.

Extract word embedding

Please download cc.en.300.bin and BioWordVec_PubMed_MIMICIII_d200.bin and run python word_embed.py to generate required json files. You need to change the path of word embedding.

Reproduce

sagri

bash let_train.sh

The results will be saved in the folder runs. You can use flag --continue_from path_to_checkpoint to continue training from a checkpoint for further training strategy exploration.