Spaces:
Runtime error
Runtime error
File size: 3,195 Bytes
217e2d4 e9007ce a1f805b 217e2d4 a1f805b 018d244 e9007ce |
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 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 |
---
title: Comment Sentiment and Toxicity Classifier
emoji: 📝
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 5.27.0
app_file: app.py
pinned: false
---
# Comment Sentiment and Toxicity Classifier
This Space utilises a custom fine-tuned BERT model to classify the sentiment and toxicity of comments. Developed for academic purposes in Australia.
# Comment MTL BERT Model
This is a BERT-based multi-task learning model capable of performing sentiment analysis and toxicity detection simultaneously.
## Model Architecture
The model is based on the `bert-base-uncased` pre-trained model with two separate classification heads:
- **Sentiment Analysis Head**: 3-class classification (Negative, Neutral, Positive)
- **Toxicity Detection Head**: 6-class multi-label classification (toxic, severe_toxic, obscene, threat, insult, identity_hate)
### Technical Parameters
- Hidden size: 768
- Number of attention heads: 12
- Number of hidden layers: 12
- Vocabulary size: 30522
- Maximum position embeddings: 512
- Hidden activation function: gelu
- Dropout probability: 0.1
## Usage
### Loading the Model
```python
from transformers import AutoTokenizer
from src.model import CommentMTLModel
import torch
# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
# Load model
model = CommentMTLModel(
model_name="bert-base-uncased",
num_sentiment_labels=3,
num_toxicity_labels=6
)
# Load pre-trained weights
state_dict = torch.load("model.bin", map_location=torch.device('cpu'))
model.load_state_dict(state_dict)
model.eval()
```
### Model Inference
```python
# Prepare input
text = "This is a test comment."
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=128)
# Model inference
with torch.no_grad():
outputs = model(input_ids=inputs["input_ids"], attention_mask=inputs["attention_mask"])
# Get results
sentiment_logits = outputs["sentiment_logits"]
toxicity_logits = outputs["toxicity_logits"]
# Process sentiment analysis results
sentiment_probs = torch.softmax(sentiment_logits, dim=1)
sentiment_labels = {0: "Negative", 1: "Neutral", 2: "Positive"}
sentiment_prediction = sentiment_labels[sentiment_probs.argmax().item()]
# Process toxicity detection results
toxicity_probs = torch.sigmoid(toxicity_logits)
toxicity_cols = ["toxic", "severe_toxic", "obscene", "threat", "insult", "identity_hate"]
toxicity_results = {label: prob.item() for label, prob in zip(toxicity_cols, toxicity_probs[0])}
print(f"Sentiment: {sentiment_prediction}")
print(f"Toxicity probabilities: {toxicity_results}")
```
## Limitations
- This model was trained on English data only and is not suitable for other languages.
- The toxicity detection may produce false positives or negatives in edge cases.
- The model may lose information when processing long texts as the maximum input length is limited to 128 tokens.
## Citation
If you use this model, please cite our repository:
```
@misc{comment-mtl-bert,
author = {Aseem},
title = {Comment MTL BERT: Multi-Task Learning for Comment Analysis},
year = {2023},
publisher = {GitHub},
url = {https://huggingface.co/Aseemks07/comment_mtl_bert_best}
}
``` |