Dual-engine AI that classifies news as Real or Fake and predicts whether it will go Viral or Not — powered by BERT, BiLSTM, and ensemble ML models trained on 100K+ articles.
Paste any news article, headline, or text below. Our AI will analyze it instantly.
⚠️ URL scraping requires backend connectivity. For demo, use text input.
💡 Headline-only analysis uses title-based features. Full text gives better results.
A multi-stage AI pipeline processes your news through NLP preprocessing, feature extraction, and dual-model inference.
Text is cleaned, tokenized, stop-words removed, and lemmatized using NLTK. URLs, special characters, and noise are stripped.
TF-IDF vectorization, Word2Vec embeddings, sentiment scores, readability metrics, and 58 statistical features are computed.
BERT/BiLSTM classifies authenticity. XGBoost/Random Forest predicts popularity score based on content and engagement features.
Confidence scores, feature importance, and SHAP-based explanations are returned with a detailed breakdown of the decision.
State-of-the-art models trained on benchmark datasets used by top research institutions worldwide.
Bidirectional Encoder Representations from Transformers
Pre-trained transformer model fine-tuned on fake news datasets. Captures deep contextual semantics bidirectionally.
Bidirectional Long Short-Term Memory
Processes text in both forward and backward directions, capturing long-range dependencies in news articles.
Ensemble Decision Tree Classifier
Ensemble of decision trees with TF-IDF features. Fast, interpretable, and robust against overfitting.
Extreme Gradient Boosting
Gradient boosting for popularity prediction. Uses 58 engineered features including sentiment, readability, and metadata.
Baseline Linear Classifier
High-performing baseline with TF-IDF features. Surprisingly competitive and highly interpretable.
Multi-Model Consensus System
Combines BERT, BiLSTM, and Random Forest via soft voting for maximum robustness across diverse news types.
44,898 articles from Reuters (real) and flagged unreliable sources (fake). Gold standard benchmark.
44,898 articles12,836 short statements from PolitiFact with 6-class labels from pants-fire to true.
12,836 statementsPolitiFact + GossipCop news with social context, user engagement, and knowledge graph data.
23,000+ articles39,797 Mashable articles with 58 features for predicting social media shares and virality.
39,797 articlesSimultaneously detects fake news AND predicts popularity — two models, one seamless analysis pipeline.
Sub-second inference with optimized model serving via Flask REST API.
SHAP values and feature importance scores explain every prediction transparently.
From sentiment polarity and readability scores to keyword density, title subjectivity, and publication timing — every signal matters.
Politics, health, science, entertainment — trained across all news categories.
Every result comes with a calibrated confidence percentage, not just a binary label.