divyanshu-chauhan-7786/fake-news-roberta

TruthLens – Fake News Detection Model (RoBERTa)

TruthLens is an advanced RoBERTa-based Fake News Detection Transformer model fine-tuned by Divyanshu Chauhan.
This model is capable of analyzing news articles and social media text and classifying them as REAL or FAKE along with a confidence score.

It is trained on the WELFake Dataset containing over 72,000 labeled news samples, balanced and cleaned for high-accuracy text classification.

This model powers the TruthLens Flask Application, capable of real-time misinformation detection.

---

🧠 Model Details

Model Developer

• Developed by: Divyanshu Chauhan
• Specialization: Artificial Intelligence & Machine Learning
• Project Purpose: Real-time misinformation detection using modern NLP

Model Type

• Architecture: RoBERTa-base (Transformer)
• Task: Binary Text Classification (Fake vs Real)
• Framework: PyTorch + HuggingFace Transformers
• Fine-tuned Model: This model

Model Capabilities

• Detects whether a news text is REAL or FAKE
• Provides a probability/confidence score
• Works on short and long news text
• Handles social media misinformation patterns

---

Training Details

✔ Dataset

• Name: WELFake
• Samples: ~72,000
• Balanced: Yes (after synthetic balancing)
• Labels:
  • 0 = Fake
  • 1 = Real

Data Preprocessing Performed

• Removed null entries
• Merged title + text → combined_text
• Cleaned URLs, numbers, special characters
• Lowercasing and stopword removal
• Train/Test split
• Used original text for Transformer fine-tuning (Transformers do their own tokenization)

Training Setup

• Base Model: roberta-base
• Epochs: 3
• Max Length: 256 tokens
• Batch Size: 8
• Learning Rate: 2e-5
• Optimizer: AdamW
• Loss: Cross-Entropy
• Hardware: GPU (Google Colab / Tesla T4)

Model Performance

Metric	Score
Accuracy	~95%
F1 Score	~95%
Precision	~95%
Recall	~95%

This makes the model suitable for production-level applications.

---

Usage Example

from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline

model_path = "divyanshu-chauhan-7786/fake-news-roberta"

tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForSequenceClassification.from_pretrained(model_path)

clf = pipeline("text-classification", model=model, tokenizer=tokenizer)

text = "Breaking: Government announces new education reforms!"
result = clf(text, truncation=True, max_length=256)

print(result)