Update README.md
Browse files
README.md
CHANGED
@@ -3,17 +3,26 @@ license: mit
|
|
3 |
---
|
4 |
<h1>Transformer Encoder for Social Science (TESS)</h1>
|
5 |
|
6 |
-
|
7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
<h2>Training Corpus</h2>
|
10 |
|
11 |
| TEXT | SOURCE |
|
12 |
| ------------- | ------------- |
|
13 |
| Preferential Trade Agreements | ToTA |
|
14 |
-
| Congressional Bills |
|
15 |
|UNGA Resolutions | UN |
|
16 |
|Firms' Annual Reports | Loughran and McDonald (2016)|
|
17 |
| U.S. Court Opinions | Caselaw Access Project|
|
18 |
|
19 |
The model is trained on 4 NVIDIA A100 GPUs for 120K steps.
|
|
|
|
3 |
---
|
4 |
<h1>Transformer Encoder for Social Science (TESS)</h1>
|
5 |
|
6 |
+
# Transformer Encoder for Social Science (TESS)
|
7 |
|
8 |
+
TESS is a deep neural network model intended for social science related NLP tasks. The model is developed by Haosen Ge, In Young Park, Xuancheng Qian, and Grace Zeng.
|
9 |
+
|
10 |
+
We demonstrate in two validation tests that TESS outperforms BERT and RoBERTa by 16.7\% on average, especially when the number of training samples is limited (<1,000 training instances). The results display the superiority of TESS on social science text processing tasks.
|
11 |
+
|
12 |
+
The pretrained model weights can be found on Hugging Face: [TESS_768_v1](https://huggingface.co/hsge/TESS_768_v1).
|
13 |
+
|
14 |
+
Working paper coming soon ...
|
15 |
|
16 |
<h2>Training Corpus</h2>
|
17 |
|
18 |
| TEXT | SOURCE |
|
19 |
| ------------- | ------------- |
|
20 |
| Preferential Trade Agreements | ToTA |
|
21 |
+
| Congressional Bills | Kornilova and Eidelman (2019) |
|
22 |
|UNGA Resolutions | UN |
|
23 |
|Firms' Annual Reports | Loughran and McDonald (2016)|
|
24 |
| U.S. Court Opinions | Caselaw Access Project|
|
25 |
|
26 |
The model is trained on 4 NVIDIA A100 GPUs for 120K steps.
|
27 |
+
|
28 |
+
|