🚀 Cyberbullying Detection in Texts, Images, and Audios

Detecting cyberbullying in online content using Machine Learning & Deep Learning techniques.

---

🌜 Table of Contents

• Introduction
• Features
• Installation
• Usage
• Datasets
• Evaluation Metrics
• Results
• Conclusion
• Future Work
• References

---

📝 Introduction

Cyberbullying is a growing concern in social media and online communities. This project leverages Machine Learning & Deep Learning models to detect cyberbullying across multiple data formats:

• Text using Bidirectional LSTM (Bi-LSTM)
• Images using Convolutional Neural Networks (CNN)
• Audio using Speech Recognition and NLP

The project aims to provide a high-accuracy cyberbullying detection system to ensure a safer online environment.

---

🌟 Features

✅ Detection of cyberbullying in text, images, and audio
✅ Ensemble approach combining Bi-LSTM and CNN for better accuracy
✅ Multi-modal data support
✅ Performance evaluation using accuracy, precision, recall, and confusion matrix
✅ Potential extension to video-based analysis

---

⚙️ Installation

Clone the repository and install dependencies:

git clone https://github.com/saiteja007-mv/Cyberbully-Detection-in-Texts-Images-and-Audios.git
cd Cyberbully-Detection-in-Texts-Images-and-Audios
pip install -r requirements.txt

Make sure you have Jupyter Notebook installed:

pip install notebook

---

🚀 Usage

1️⃣ Open Jupyter Notebook

jupyter notebook

2️⃣ Navigate to the notebook files for text, image, and audio analysis
3️⃣ Prepare your dataset with labeled cyberbullying and non-cyberbullying content
4️⃣ Run the cells in the notebook to train the models
5️⃣ View evaluation metrics and predictions

---

📂 Datasets

📝 Text Dataset

• File: twitter.csv
• Content: Labeled tweets for cyberbullying detection
  

🖼️ Images Dataset

• Folder: Images_dataset
• Content: Cyberbullying-related images
  

🔊 Audio Dataset

• Folder: audios
• Content: Audio samples labeled for cyberbullying
  

---

📊 Evaluation Metrics

The models are evaluated based on:

• 💢 Accuracy: Percentage of correctly classified instances
• 📍 Precision: Ratio of correctly predicted positive observations
• 📋 Recall: Proportion of actual positives correctly identified
• 📊 Confusion Matrix: Visual representation of true vs. predicted labels

---

📈 Results

• High accuracy achieved for text and image-based cyberbullying detection
• The ensemble approach (Bi-LSTM + CNN) improved precision & recall
• Future improvements can enhance audio analysis

---

🎯 Conclusion

This project demonstrates that deep learning models can effectively detect cyberbullying across different media types. The ensemble approach combining Bidirectional LSTM and CNN significantly improves accuracy. However, further enhancements are required for real-time video analysis.

---

🔮 Future Work

🔹 Integrate video analysis using object recognition and context analysis
🔹 Enhance speech recognition for better audio-based detection
🔹 Deploy as an API or web application for real-time cyberbullying detection
🔹 Improve dataset diversity to enhance model generalization

---

📚 References

📖 V. Subrahmanian and S. Kumar, Predicting human behavior: The next frontiers, Science, vol. 355, no. 6324, p. 489, 2017.

📖 H. Lauw, J. C. Shafer, R. Agrawal, and A. Ntoulas, Homophily in the digital world: A Live Journal case study, IEEE Internet Comput., vol. 14, no.2, pp. 15–23, 2010.

📖 Arindam Mandal, Sriparna Saha, Alexandar Ferworn, A Hybrid Model for Cyberbullying Detection on Facebook.