Update README.md
Browse files
README.md
CHANGED
@@ -5,7 +5,7 @@ license: apache-2.0
|
|
5 |
datasets:
|
6 |
- tweets
|
7 |
widget:
|
8 |
-
- text: "COVID-19
|
9 |
---
|
10 |
|
11 |
# Disclaimer: This page is under maintenance. Please DO NOT refer to the information on this page to make any decision yet.
|
@@ -17,6 +17,14 @@ Fine-tuned model on English language using a masked language modeling (MLM) obje
|
|
17 |
|
18 |
## Model description
|
19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
You can embed local or remote images using `![](...)`
|
21 |
|
22 |
## Intended uses & limitations
|
@@ -33,6 +41,18 @@ Provide examples of latent issues and potential remediations.
|
|
33 |
|
34 |
## Training data
|
35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
Describe the data you used to train the model.
|
37 |
If you initialized it with pre-trained weights, add a link to the pre-trained model card or repository with description of the pre-training data.
|
38 |
|
|
|
5 |
datasets:
|
6 |
- tweets
|
7 |
widget:
|
8 |
+
- text: "COVID-19 vaccines are safe and effective."
|
9 |
---
|
10 |
|
11 |
# Disclaimer: This page is under maintenance. Please DO NOT refer to the information on this page to make any decision yet.
|
|
|
17 |
|
18 |
## Model description
|
19 |
|
20 |
+
- Baseline model: BERTweet14,15
|
21 |
+
- trained based on the RoBERTa pre-training procedure
|
22 |
+
- 850M General English Tweets (Jan 2012 ~ Aug 2019)
|
23 |
+
- 23M COVID-19 English Tweets
|
24 |
+
- Size of the model: >134M parameters
|
25 |
+
- Further training
|
26 |
+
- Training with recent COVID-19 and vaccine tweets
|
27 |
+
|
28 |
You can embed local or remote images using `![](...)`
|
29 |
|
30 |
## Intended uses & limitations
|
|
|
41 |
|
42 |
## Training data
|
43 |
|
44 |
+
#### 1) Pre-training language model
|
45 |
+
- Tweets with trending #CovidVaccine hashtag 207,000 tweets uploaded across 2020-08-18 ~ 2021-04-20 [3]
|
46 |
+
- Tweets about all COVID-19 vaccines 78,000 tweets uploaded across 2020-12-20 ~ 2021-05-13 [4]
|
47 |
+
- Covid-19 Twitter chatter dataset 590,000 tweets uploaded across 2021-03-01 ~ 2021-05-20 [5]
|
48 |
+
|
49 |
+
#### 2) Fine-tuning for fact classification
|
50 |
+
- Statements from Poynter and Snopes with Selenium 14,000 fact-checked statements from 2020-01-14 to 2021-05-13
|
51 |
+
- Divide original labels within 3 categories
|
52 |
+
False: False, no evidence, manipulated, fake, not true, unproven, unverified
|
53 |
+
Misleading: Misleading, exaggerated, out of context, needs context
|
54 |
+
True: True, correct
|
55 |
+
|
56 |
Describe the data you used to train the model.
|
57 |
If you initialized it with pre-trained weights, add a link to the pre-trained model card or repository with description of the pre-training data.
|
58 |
|