Update README.md
Browse files
README.md
CHANGED
|
@@ -17,6 +17,9 @@ tags:
|
|
| 17 |
# Model Details
|
| 18 |
Political DEBATE (DeBERTa Algorithm for Textual Entailment) is an NLI classifier trained for zero-shot and few-shot classification of political documents.
|
| 19 |
|
|
|
|
|
|
|
|
|
|
| 20 |
The model was trained using [Moritz Laurer's](https://huggingface.co/MoritzLaurer/deberta-v3-large-zeroshot-v2.0-c) zero shot model as the foundation, and then trained further using the [PolNLI dataset.](https://huggingface.co/datasets/mlburnham/Pol_NLI) The model was trained for four classification categories:
|
| 21 |
1. Stance detection (i.e. opinion mining).
|
| 22 |
2. Hatespeech detection.
|
|
@@ -27,10 +30,10 @@ The PolNLI dataset contains documents from social media, news outlets, congressi
|
|
| 27 |
|
| 28 |
If you use this model or the PolNLI data set please cite:
|
| 29 |
```
|
| 30 |
-
@article{
|
| 31 |
title={Political DEBATE: Efficient Zero-shot and Few-shot Classifiers for Political Text},
|
| 32 |
-
author={Burnham, Michael, Kayla
|
| 33 |
-
journal={
|
| 34 |
year={2024}
|
| 35 |
}
|
| 36 |
```
|
|
@@ -62,7 +65,7 @@ Evaluation is primarily conducted on the PolNLI test set. No hypotheses in the P
|
|
| 62 |
<img src="https://cdn-uploads.huggingface.co/production/uploads/64d0341901931c60161f2a06/-sQhURg-zeacZAUNEVUpc.png" width="750" height="500" />
|
| 63 |
|
| 64 |
### Performance on the PolNLI test set by task:
|
| 65 |
-
<img src="https://cdn-uploads.huggingface.co/production/uploads/64d0341901931c60161f2a06/
|
| 66 |
|
| 67 |
### Few-shot performance on open-text survey answers vs. Llama 3.1 8B (unquantized):
|
| 68 |
<img src="https://cdn-uploads.huggingface.co/production/uploads/64d0341901931c60161f2a06/kDZXC-OsZtmTCxq0HsJMg.png" width="750" height="500" />
|
|
|
|
| 17 |
# Model Details
|
| 18 |
Political DEBATE (DeBERTa Algorithm for Textual Entailment) is an NLI classifier trained for zero-shot and few-shot classification of political documents.
|
| 19 |
|
| 20 |
+
[Zero-shot tutorial](https://colab.research.google.com/drive/1zi-8pMx_x-vo0m8XVmYVtw0Dhdw7vfFD#scrollTo=T4bo1aEjB9iG)
|
| 21 |
+
[Few-shot tutorial](https://colab.research.google.com/drive/1Sv82jqRSwiIyuvEIDrhTiqF8_ClQaL0r#scrollTo=Uuaw8-qpn8S6)
|
| 22 |
+
|
| 23 |
The model was trained using [Moritz Laurer's](https://huggingface.co/MoritzLaurer/deberta-v3-large-zeroshot-v2.0-c) zero shot model as the foundation, and then trained further using the [PolNLI dataset.](https://huggingface.co/datasets/mlburnham/Pol_NLI) The model was trained for four classification categories:
|
| 24 |
1. Stance detection (i.e. opinion mining).
|
| 25 |
2. Hatespeech detection.
|
|
|
|
| 30 |
|
| 31 |
If you use this model or the PolNLI data set please cite:
|
| 32 |
```
|
| 33 |
+
@article{burnham2024political,
|
| 34 |
title={Political DEBATE: Efficient Zero-shot and Few-shot Classifiers for Political Text},
|
| 35 |
+
author={Burnham, Michael and Kahn, Kayla and Wang, Ryan Yank and Peng, Rachel X},
|
| 36 |
+
journal={arXiv preprint arXiv:2409.02078},
|
| 37 |
year={2024}
|
| 38 |
}
|
| 39 |
```
|
|
|
|
| 65 |
<img src="https://cdn-uploads.huggingface.co/production/uploads/64d0341901931c60161f2a06/-sQhURg-zeacZAUNEVUpc.png" width="750" height="500" />
|
| 66 |
|
| 67 |
### Performance on the PolNLI test set by task:
|
| 68 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/64d0341901931c60161f2a06/ss6cN8ujGl40SZ5WxjiDH.png" width="750" height="500" />
|
| 69 |
|
| 70 |
### Few-shot performance on open-text survey answers vs. Llama 3.1 8B (unquantized):
|
| 71 |
<img src="https://cdn-uploads.huggingface.co/production/uploads/64d0341901931c60161f2a06/kDZXC-OsZtmTCxq0HsJMg.png" width="750" height="500" />
|