|
|
--- |
|
|
license: mit |
|
|
datasets: |
|
|
- thesofakillers/jigsaw-toxic-comment-classification-challenge |
|
|
language: |
|
|
- en |
|
|
metrics: |
|
|
- accuracy |
|
|
- f1 |
|
|
tags: |
|
|
- text-classification |
|
|
- toxic_comment |
|
|
- nlp |
|
|
- transformers |
|
|
- distilbert |
|
|
pipeline_tag: text-classification |
|
|
--- |
|
|
# Toxic Comment Classifier (Distil-bert-uncased) |
|
|
|
|
|
This model is a fine-tuned **Distil-bert-uncased** model for **toxic comment classification**. |
|
|
It classifies comments as either **toxic** or **non-toxic**. |
|
|
|
|
|
|
|
|
## Training |
|
|
|
|
|
The model was trained using Hugging Face `Trainer` on a labeled toxic comment dataset. |
|
|
Evaluation metrics: |
|
|
|
|
|
- **Accuracy:** ~97% |
|
|
- **F1 score:** ~83% |
|
|
|
|
|
## Intended Use |
|
|
|
|
|
- Detecting toxic or harmful language in text. |
|
|
- Useful for moderation in forums, social media, and chat systems. |
|
|
|
|
|
## Limitations |
|
|
|
|
|
- May not capture sarcasm or subtle toxicity. |
|
|
- Biases in the training dataset may affect predictions. |
|
|
|
|
|
## Usage |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline |
|
|
|
|
|
model_id = "Youssef-El-SaYed/toxic-comment-classifier" |
|
|
|
|
|
# Define mapping |
|
|
id2label = {0: "Non-Toxic", 1: "Toxic"} |
|
|
label2id = {"Non-Toxic": 0, "Toxic": 1} |
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(model_id) |
|
|
model = AutoModelForSequenceClassification.from_pretrained( |
|
|
model_id, |
|
|
id2label=id2label, |
|
|
label2id=label2id |
|
|
) |
|
|
|
|
|
nlp = pipeline("text-classification", model=model, tokenizer=tokenizer) |
|
|
|
|
|
print(nlp("You are so stupid and annoying!")) |
|
|
print(nlp("I really like your work, keep it up!")) |
|
|
|
|
|
|
|
|
|
