You can deploy this model and start earning money today!

Sparker596/
clickbait-classification

Name: clickbait-classification
Availability: InStock

safetensors

distilbert

Views

No views yet

Model Card: DistilBERT Clickbait Title Classifier

Model Description

This model is a fine-tuned version of distilbert-base-uncased for binary classification of clickbait titles. It predicts whether a given title is clickbait (1) or not clickbait (0).

Model type: Text Classification (Binary)
Base model: distilbert/distilbert-base-uncased
Language: English
License: Apache 2.0 (inherited from base model)

Intended Use

Primary Use Cases

Detecting clickbait headlines in news articles, blog posts, and social media content
Content moderation and quality filtering
Research on sensationalist content patterns

Out-of-Scope Uses

This model should not be used as the sole arbiter for content removal decisions
Not suitable for languages other than English
Not designed for nuanced editorial judgment or content quality assessment beyond clickbait detection

Training Data

Dataset: marksverdhei/clickbait_title_classification
Training split: 80% of original training data (further split into 80% train, 20% validation)
Test split: 20% of original training data

The dataset contains titles labeled as clickbait or non-clickbait. The model was trained on text titles only, with the label column renamed to 'labels' for binary classification.

Training Procedure

Preprocessing

Tokenization using DistilBERT tokenizer (uncased)
Dynamic padding with DataCollatorWithPadding
Maximum sequence length: 512 tokens (inference)
Batched tokenization with batch size of 32

Training Hyperparameters

Epochs: 3
Batch size: 32 (train and eval)
Evaluation strategy: Every 500 steps
Logging: Every 200 steps
Checkpointing: Every 1000 steps
Optimizer: AdamW (default)
Best model selection: Based on validation accuracy

Training Infrastructure

Framework: Hugging Face Transformers
Training approach: Supervised fine-tuning

Performance

The model achieved excellent performance on the validation set:

Metric	Score
Accuracy	99.19%
Precision	99.61%
Recall	98.72%
F1 Score	99.17%

Training Progress

Step	Training Loss	Validation Loss	Accuracy	Precision	Recall	F1
500	0.0623	0.0516	98.56%	97.74%	99.36%	98.54%
1000	0.0114	0.0412	99.11%	99.26%	98.91%	99.09%
1500	0.0133	0.0386	99.19%	99.61%	98.72%	99.17%
2000	0.0025	0.0511	98.95%	98.45%	99.43%	98.94%

Best checkpoint: Step 1500 (selected based on highest validation accuracy)

Usage

Inference Example

python

import torch
import numpy as np
from transformers import AutoTokenizer, AutoModelForSequenceClassification

# Load model and tokenizer
model = AutoModelForSequenceClassification.from_pretrained("path/to/model")
tokenizer = AutoTokenizer.from_pretrained("path/to/model")

# Prepare model
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
model.eval()

# Inference function
def predict_clickbait(text):
    with torch.no_grad():
        inputs = tokenizer(text, return_tensors='pt', truncation=True, 
                          padding=True, max_length=512)
        inputs = {k: v.to(device) for k, v in inputs.items()}
        outputs = model(**inputs)
        probabilities = torch.nn.functional.softmax(outputs.logits, dim=-1).cpu().numpy()
        predicted_class = np.argmax(probabilities, axis=1).item()
    
    return {
        'prediction': 'clickbait' if predicted_class == 1 else 'not clickbait',
        'confidence': probabilities[0][predicted_class]
    }

# Example usage
result = predict_clickbait("You Won't Believe What Happened Next!")
print(result)

Limitations

Domain specificity: Trained on a specific clickbait dataset; performance may vary on different types of content
Language: English only
Potential overfitting: Very high validation scores (>99%) suggest possible overfitting to the validation set
Context-independent: Only evaluates titles in isolation, without considering article content
Temporal drift: Clickbait patterns evolve over time; model may need retraining on newer data
Binary classification: Does not provide nuanced assessment of clickbait severity

Ethical Considerations

Bias

The model reflects biases present in the training data
May exhibit different performance across content domains, publication types, or cultural contexts
Could potentially misclassify legitimate attention-grabbing headlines as clickbait

Fairness

Should not be used to unfairly penalize specific publishers or content creators
Decisions based on this model should allow for human review and appeal

Privacy

The model processes text only and does not collect user data
No personal information is used in training or inference

Citation

If you use this model, please cite:

bibtex

@misc{clickbait-distilbert-classifier,
  author = {[Your Name]},
  title = {DistilBERT Clickbait Title Classifier},
  year = {2025},
  publisher = {[Platform]},
  note = {Fine-tuned on marksverdhei/clickbait_title_classification}
}

Model Card Authors

[]

Model Card Contact

[madhajaswanth@gmail.com]

Last updated: January 2025

Sparker596/clickbait-classification