Sarcasm Detection using Hierarchical BERT + BiLSTM + CNN

This project implements a sarcasm detection system using a hybrid NLP architecture combining BERT, BiLSTM, and CNN layers. It identifies whether a given news headline is sarcastic or not.

📁 Dataset

• Dataset: Sarcasm_Headlines_Dataset.json

🧪 Model Architecture

• BERT Tokenizer for contextual token embeddings
• TFBertModel from Hugging Face as base
• BiLSTM layer to capture sequence dependencies
• CNN layers to extract n-gram level features
• Dense layer for binary classification

📊 Performance

• Trained on 10,000 samples
• Robust at detecting sarcastic patterns

🛠️ Tech Stack

• Python
• TensorFlow + Keras
• Hugging Face Transformers
• Pandas, NumPy
• Streamlit (for deployment)