This project applies deep learning to classify news articles as real or fake, leveraging CNN-Bidirectional LSTM for advanced feature extraction and sequence learning. The model is trained on labeled news data to mitigate misinformation spread on digital platforms.
-
Data Collection & Preprocessing
- Uses 'train.tsv' for training and 'test.tsv' for evaluation.
- Target variable: 'label' (1: Fake, 0: Real).
- Text preprocessing includes stopword removal and token filtering using NLTK & Gensim.
- Tokenization & padding ensure uniform input size.
-
Model Architecture
- Embedding Layer: Converts words into dense vector representations.
- Convolutional Layers: Extracts local text features using Conv1D & MaxPooling.
- Bidirectional LSTM: Captures context from both past and future sequences.
- Dropout Layer: Prevents overfitting.
- Dense Layer: Outputs the probability of a news article being fake or real.
-
Training & Evaluation
- Loss Function: Binary Cross-Entropy.
- Optimizer: Adam.
- Metrics: Accuracy, Precision, Recall, F1 Score, ROC AUC.
- Training Configuration:
- 5 epochs, batch size 64.
- 80-20 train-validation split.
- Achieved 97% accuracy in detecting fake news.
-
Future Work
- Extend to multilingual datasets.
- Deploy model for real-time news classification.
Ensure you have the required dependencies installed before running the code:
pip install pandas numpy tensorflow nltk gensim matplotlib- Clone this repository:
git clone https://github.com/your-repo/news-classification.git cd news-classification
- Model Performance Metrics:
- Accuracy: 97%
- Precision: 97%
- Recall: 97%
- F1 Score: 97%
- ROC AUC Score: 97%
- Strong generalization on unseen data.
- Sachin Gaikwad
- Aarya Pawar
- Shikhar Kanauje
- Yash Dilip Phalke
For any queries, feel free to open an issue.