GitHub - sfschouten/entity-news

This repository contains code to perform various experiments involving named entities.

dataset_mwep.py: loads dataset collected with MWEP, in particular the RefNews-12 dataset.
dataset_el_wiki.py wrapper around KILT Wikipedia for use with entity linking.

Probing the representations of named entities in Transformer-based Language Models @ BlackboxNLP 2022

The experimental results of this paper were generated using this code base.

Entity substitution experiment:

Both these files output the logits for their predictions, these are read by the analyse_uncertainty.py file to obtain the uncertainty estimates.

The diagnostic classifiers were trained with train_ner.py by using the --probing option.

Name		Name	Last commit message	Last commit date
Latest commit History 137 Commits
.gitignore		.gitignore
README.md		README.md
analyse_uncertainty.py		analyse_uncertainty.py
data_collator.py		data_collator.py
dataset_el_wiki.py		dataset_el_wiki.py
dataset_mwep.py		dataset_mwep.py
environment.yaml		environment.yaml
experiment_confident_mistakes.py		experiment_confident_mistakes.py
experiment_entitypoor_mlm.py		experiment_entitypoor_mlm.py
experiment_entitypoor_news_clf.py		experiment_entitypoor_news_clf.py
experiment_nerc_with_analysis.py		experiment_nerc_with_analysis.py
experiment_visualize_entity_tokens.py		experiment_visualize_entity_tokens.py
modeling_entity_linking.py		modeling_entity_linking.py
modeling_versatile.py		modeling_versatile.py
multitask_trainer.py		multitask_trainer.py
train_mlm.py		train_mlm.py
train_nel.py		train_nel.py
train_ner.py		train_ner.py
train_nerc.py		train_nerc.py
train_news_clf+ner.py		train_news_clf+ner.py
train_news_clf.py		train_news_clf.py
trainer.py		trainer.py
utils.py		utils.py
utils_dataset_enrich.py		utils_dataset_enrich.py
utils_mentions.py		utils_mentions.py