Update README.md
Browse files
README.md
CHANGED
|
@@ -8,4 +8,45 @@ license: apache-2.0
|
|
| 8 |
|
| 9 |
---
|
| 10 |
|
| 11 |
-
# RuBERTConv Toxic Classifier
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
---
|
| 10 |
|
| 11 |
+
# RuBERTConv Toxic Classifier
|
| 12 |
+
|
| 13 |
+
## Model description
|
| 14 |
+
|
| 15 |
+
Based on [rubert-base-cased-conversational](https://huggingface.co/DeepPavlov/rubert-base-cased-conversational) model
|
| 16 |
+
|
| 17 |
+
## Intended uses & limitations
|
| 18 |
+
|
| 19 |
+
#### How to use
|
| 20 |
+
|
| 21 |
+
```python
|
| 22 |
+
from transformers import pipeline
|
| 23 |
+
|
| 24 |
+
model_name = "MindfulSquirrel/rubertconv_toxic_clf"
|
| 25 |
+
pipe = pipeline("text-classification", model=model_name, tokenizer=model_name, framework="pt")
|
| 26 |
+
|
| 27 |
+
text = "Ты придурок из интернета"
|
| 28 |
+
pipe([text])
|
| 29 |
+
```
|
| 30 |
+
|
| 31 |
+
## Training data
|
| 32 |
+
|
| 33 |
+
Datasets:
|
| 34 |
+
- [2ch]( https://www.kaggle.com/blackmoon/russian-language-toxic-comments)
|
| 35 |
+
- [Odnoklassniki](https://www.kaggle.com/alexandersemiletov/toxic-russian-comments)
|
| 36 |
+
- [Toloka Persona Chat Rus](https://toloka.ai/ru/datasets)
|
| 37 |
+
- [Koziev's Conversations](https://github.com/Koziev/NLP_Datasets/blob/master/Conversations/Data) with [toxic words vocabulary](https://www.dropbox.com/s/ou6lx03b10yhrfl/bad_vocab.txt.tar.gz)
|
| 38 |
+
|
| 39 |
+
Augmentations:
|
| 40 |
+
- ё -> е
|
| 41 |
+
- Remove or add "?" or "!"
|
| 42 |
+
- Fix CAPS
|
| 43 |
+
- Concatenate toxic and non-toxic texts
|
| 44 |
+
- Concatenate two non-toxic texts
|
| 45 |
+
- Add toxic words from vocabulary
|
| 46 |
+
- Add typos
|
| 47 |
+
- Mask toxic words with "*", "@", "$"
|
| 48 |
+
|
| 49 |
+
|
| 50 |
+
## Training procedure
|
| 51 |
+
|
| 52 |
+
TBA
|