Add more info to readme
Browse files
README.md
CHANGED
@@ -1,7 +1,40 @@
|
|
1 |
# German sentiment BERT finetuned on news data
|
2 |
-
|
3 |
-
Based on https://huggingface.co/oliverguhr/german-sentiment-bert, with
|
4 |
-
additional training on news data.
|
5 |
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
# German sentiment BERT finetuned on news data
|
|
|
|
|
|
|
2 |
|
3 |
+
Sentiment analysis model based on https://huggingface.co/oliverguhr/german-sentiment-bert, with additional training on German news texts about migration.
|
4 |
+
|
5 |
+
This model is part of the project https://github.com/text-analytics-20/news-sentiment-development, which explores sentiment development in German news articles about migration between 2007 and 2019.
|
6 |
+
|
7 |
+
Code for inference (predicting sentiment polarity) on raw text can be found at https://github.com/text-analytics-20/news-sentiment-development/blob/main/sentiment_analysis/bert.py
|
8 |
+
|
9 |
+
If you are not interested in polarity but just want to predict discrete class labels (0: positive, 1: negative, 2: neutral), you can also use the model with Oliver Guhr's `germansentiment` package as follows:
|
10 |
+
|
11 |
+
First install the package from PyPI:
|
12 |
+
|
13 |
+
```bash
|
14 |
+
pip install germansentiment
|
15 |
+
```
|
16 |
+
|
17 |
+
Then you can use the model in Python:
|
18 |
+
|
19 |
+
```python
|
20 |
+
from germansentiment import SentimentModel
|
21 |
+
|
22 |
+
model = SentimentModel('mdraw/german-news-sentiment-bert')
|
23 |
+
|
24 |
+
# Examples from our validation dataset
|
25 |
+
texts = [
|
26 |
+
'[...], schwärmt der parteilose Vizebürgermeister und Historiker Christian Matzka von der "tollen Helferszene".',
|
27 |
+
'Flüchtlingsheim 11.05 Uhr: Massenschlägerei',
|
28 |
+
'Rotterdam habe einen Migrantenanteil von mehr als 50 Prozent.',
|
29 |
+
]
|
30 |
+
|
31 |
+
result = model.predict_sentiment(texts)
|
32 |
+
|
33 |
+
print(result)
|
34 |
+
```
|
35 |
+
|
36 |
+
The code above will print:
|
37 |
+
|
38 |
+
```python
|
39 |
+
['positive', 'negative', 'neutral']
|
40 |
+
```
|