Welcome to WISER (Weak and Indirect Supervision for Entity Recognition), a system for training sequence tagging models, particularly neural networks for named entity recognition (NER) and related tasks. WISER uses weak supervision in the form of rules to train these models, as opposed to hand-labeled training data.
The original WISER paper can be accessed here.
| Method | NCBI-Disease (F1) | BC5CDR (F1) | LaptopReview (F1) |
|---|---|---|---|
| AutoNER | 75.52 | 82.13 | 65.44 |
| Snorkel | 73.41 | 82.24 | 63.54 |
| WISER | 79.03 | 82.94 | 69.04 |
These instructions will WISER up and running on your local machine to develop your own pipelines for weakly supervised for sequence tagging tasks.
WISER requires Python 3.7. To install the required dependencies, please run
pip install -r requirements.txt
Or alternatively
conda install --file requirements.txt
Then, inside the wiser directory, please run
pip install .
Refer to tutorial/introduction for a comprehensive introduction to using WISER to train end-to-end frameworks with weak supervision. More tutorials coming soon!
Once you're comfortable with the WISER framework, we recommend looking at our FAQ for strategies on how to write rules and debug your pipeline.
Please cite the following paper if you are using our tool. Thank you!
Esteban Safranchik, Shiying Luo, Stephen H. Bach. "Weakly Supervised Sequence Tagging From Noisy Rules". In 34th AAAI Conference on Artificial Intelligence, 2020.
@inproceedings{safranchik2020weakly,
title = {Weakly Supervised Sequence Tagging From Noisy Rules},
author = {Safranchik, Esteban and Luo, Shiying and Bach, Stephen H.},
booktitle = {AAAI},
year = 2020,
}