🚨 Hate Speech Detection (BERT + CNN + LSTM)

This project implements a hybrid deep learning model for detecting hate speech, offensive language, and neutral text in tweets. It combines BERT embeddings with CNN and LSTM layers to achieve high accuracy in text classification. Hosted on: Streamlit Cloud

📌 Features

• Classifies tweets/text into:
  • Hate Speech
  • Offensive Language
  • Neutral
• Hybrid model architecture:
  • BERT for embeddings
  • CNN for feature extraction
  • LSTM for sequence learning
• Streamlit Web App for real-time predictions
• Color-coded probability bars for better visualization

🗂️ Project Structure

HATE_SPEECH_DETECTION-LOCAL/
├── app/
│ ├── pycache/
│ ├── .env
│ ├── .venv/
│ ├── cnn_lstm_hate_speech_bert/
│ ├── hsd_model.py
│ ├── labeled_data.csv
│ ├── requirements.txt
│ └── streamlit_app.py
│
├── assets/
│ ├── hate_speech.png
│ ├── neutral.png
│ └── offensive.png
│
├── docs/
│ ├── certificate.jpeg
│ ├── Code run.mkv
│ ├── Deep Learning for Hate Speech Detection in Tweets Using LSTM ...
│ ├── final_ppt_hate_speech.pptx
│ ├── hate_speech_report_final_2023.pdf
│ └── hate_speech_report_final_2323.docx
│
├── Notebook/
│ ├── bert_lstm_cnn_test.ipynb
│ └── bert_lstm_cnn.ipynb
└── README.md

⚠️ Note: The trained model (cnn_lstm_hate_speech_bert/) is too large for GitHub (>100MB). It is excluded via .gitignore. You must download or export your own SavedModel.

📸 Screenshots

Hate

Neutral

Offensive

🚀 Installation

1. Clone the repo:

   git clone https://github.com/ArponRoy7/Hate_speech_detection.git
   cd Hate_speech_detection

2. Create & activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate   # Linux/Mac
   .venv\Scripts\activate      # Windows

3. Install dependencies:

   pip install -r requirements.txt

   Ensure TensorFlow and TensorFlow-Text match:

   pip install "tensorflow==2.19.0" "tensorflow-text==2.19.0"

4. Place your trained model inside the project root:

   cnn_lstm_hate_speech_bert/
   ├── saved_model.pb
   └── variables/
   

▶️ Running the App

Start the Streamlit app:

streamlit run streamlit_app.py

📝 Example Predictions

• Input: "Who tf is this faggot"
  → Hate Speech 🟨
• Input: "One fly ass bitch"
  → Offensive 🟥
• Input: "Harry Styles is so hot"
  → Neutral 🟩

📊 Dataset

• Davidson et al. — Hate Speech and Offensive Language
• Basile et al. — Hate and Not Hate
• Zampieri et al. — Offensive and Not Offensive

Final merged dataset: 34,896 tweets
Labels: 0=Hate Speech, 1=Offensive, 2=Neutral

📈 Performance

• Training & validation accuracy plotted over epochs
• Confusion matrix on test data
• Hybrid BERT + CNN + LSTM shows superior performance compared to standalone CNN, LSTM, Bi-LSTM

👩‍🏫 Mentor

• Mrs. Monika Singh, Assistant Professor, RCCIIT

👨‍💻 Authors

• Ashif Mondal
• Diya Neogi
• Arpon Roy
• Bidisha Saha