https://github.com/tteofili/cheaper

Low Cost Entity Resolution with Transformers
https://github.com/tteofili/cheaper

data-management entity-resolution record-linkage

Last synced: about 6 hours ago
JSON representation

Low Cost Entity Resolution with Transformers

• Host: GitHub
• URL: https://github.com/tteofili/cheaper
• Owner: tteofili
• Created: 2021-01-28T07:06:32.000Z (about 4 years ago)
• Default Branch: master
• Last Pushed: 2024-06-17T15:18:50.000Z (8 months ago)
• Last Synced: 2024-06-17T17:05:46.991Z (8 months ago)
• Topics: data-management, entity-resolution, record-linkage
• Language: Jupyter Notebook
• Homepage:
• Size: 10.6 MB
• Stars: 0
• Watchers: 4
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
CheapER

=======

`CheapER` is a tool for performing Entity Resolution tasks with few labeled training samples.

`CheapER` adopts large language models within a _noisy training_ framework, in combination with _adaptive fine tuning_, _consistency training_, _adaptive softmax_ and _Monte Carlo dropout_.

![*CheapER pipeline*](cheaper.png)

# Experiments

`CheapER` requires less labeled training data with respect to SotA systems (as of early 2023) to reach the same _F1_.

![*CheapER cost on DM datasets*](dm_results.png)

Experiments on the _DeepMatcher_ datasets can be reproduced using the `eval.py` script.

# Notebooks

* [Effectiveness](https://colab.research.google.com/drive/1G0PMnt4xtrwvztjmOBTPbJmwV51ajysN#scrollTo=3sonS3GiFaE1) of [adaptive fine-tuning](https://ruder.io/recent-advances-lm-fine-tuning/) for the ER task.

* [CheapER training using 5% of the BeerAdvo-RateBeer dataset](example.ipynb) (using a DistilBert model).

# Citing CheapER

If you extend or use this work, please cite:

```

@article{teofili2023cheaper,

  title={CheapER: Low Cost Entity Resolution},

  author={Teofili, Tommaso and Firmani, Donatella and Merialdo, Paolo},

  year={2023}

}

```