File size: 2,678 Bytes
0deae3c 18082dd 0deae3c dcdeb9e 0deae3c 188c48e 0deae3c dd8a2fd bf2ec66 58752af ec1b9f9 82fe319 ec1b9f9 58752af 0deae3c d701327 ec1b9f9 d701327 951b61b d701327 951b61b d701327 0deae3c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 |
---
license: apache-2.0
widget:
- text: Din store idiot
base_model: Maltehb/aelaectra-danish-electra-small-cased
---
# Danish Offensive Text Detection based on ELECTRA-small
This model is a fine-tuned version of [Maltehb/aelaectra-danish-electra-small-cased](https://huggingface.co/Maltehb/aelaectra-danish-electra-small-cased) on a dataset consisting of approximately 5 million Facebook comments on [DR](https://dr.dk/)'s public Facebook pages. The labels have been automatically generated using weak supervision, based on the [Snorkel](https://www.snorkel.org/) framework.
The model almost achieves SOTA results while being 20x smaller, on a test set consisting of 600 Facebook comments annotated using majority vote by three annotators, of which 35.8% were labelled as offensive:
| **Model** | **Precision** | **Recall** | **F1-score** | **F2-score** |
| :-------- | :------------ | :--------- | :----------- | :----------- |
| [`alexandrainst/da-offensive-detection-base`](https://huggingface.co/alexandrainst/danoff-base) | 74.81% | **89.77%** | **81.61%** | **86.32%** |
| `alexandrainst/da-offensive-detection-small` (this) | 74.13% | 89.30% | 81.01% | 85.79% |
| [`A&ttack`](https://github.com/ogtal/A-ttack) | **97.32%** | 50.70% | 66.67% | 56.07% |
| [`alexandrainst/da-hatespeech-detection-small`](https://huggingface.co/alexandrainst/da-hatespeech-detection-small) | 86.43% | 56.28% | 68.17% | 60.50% |
| [`Guscode/DKbert-hatespeech-detection`](https://huggingface.co/Guscode/DKbert-hatespeech-detection) | 75.41% | 42.79% | 54.60% | 46.84% |
## Using the model
You can use the model simply by running the following:
```python
>>> from transformers import pipeline
>>> offensive_text_pipeline = pipeline(model="alexandrainst/da-offensive-detection-small")
>>> offensive_text_pipeline("Din store idiot")
[{'label': 'Offensive', 'score': 0.9997463822364807}]
```
Processing multiple documents at the same time can be done as follows:
```python
>>> offensive_text_pipeline(["Din store idiot", "ej hvor godt :)"])
[{'label': 'Offensive', 'score': 0.9997463822364807}, {'label': 'Not offensive', 'score': 0.9996451139450073}]
```
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 32
- eval_batch_size: 32
- gradient_accumulation_steps: 1
- total_train_batch_size: 32
- seed: 4242
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- max_steps: 500000
- fp16: True
- eval_steps: 1000
- early_stopping_patience: 100
### Framework versions
- Transformers 4.20.1
- Pytorch 1.11.0+cu113
- Datasets 2.3.2
- Tokenizers 0.12.1 |