Cyberbullying Detection Project

Overview 📖

The Cyberbullying Detection Project is dedicated to creating a powerful text classification system to identify and address cyberbullying instances in digital communication in Polish language. Employing advanced machine learning models, this project categorizes text data into cyberbullying or non-cyberbullying categories, helping to foster safer online interactions.

Dataset 📊

The project leverages a comprehensive dataset comprising various text samples, meticulously labeled as either cyberbullying or non-cyberbullying content. This dataset is instrumental for training the models to accurately discern between harmful and non-harmful online interactions.

Installation 💻

To set up this project, ensure you have Python, TensorFlow, and SpaCy installed.

• Python: Ensure Python is installed. Download Python.
• TensorFlow: Install TensorFlow for Python using pip:

pip install tensorflow

• SpaCy: Install SpaCy and download the polish language model:

pip install spacy
python -m spacy download pl_core_news_sm

Usage 🛠️

To use this project, follow these steps:

1. Prepare Your Dataset:

• Place your text data in a designated directory.
• Ensure the data is in a readable format that includes labels for training.

2. Update File Paths:

• In the respective Python scripts, ensure that file paths for data loading are correctly set.

3. Run the Training Script:

• Execute the training scripts for the chosen models, such as text_categorization.py or text_cnn_classifier.py.
• Adjust parameters like epochs and batch sizes as needed.

4. Model Evaluation:

• After training, the scripts will evaluate the models using precision, recall, and F1-score metrics.
• Optionally, use the plot_metrics function to visualize the performance of each model.

Project Structure 📁

• text_preprocessor.py: Handles the preprocessing of text data.
• text_categorization.py: Defines the UNet architecture.
• text_cnn_classifier.py: TensorFlow-based CNN model for text classification.
• cyberbullying_detection.py: Main script to orchestrate data preparation, model training, evaluation, and visualization.

Key Features 🔑

• Text Classification Models: Uses both SpaCy and TensorFlow models to classify text.
• Data Preprocessing: Includes text normalization, lemmatization, and removal of irrelevant characters.
• Model Evaluation: Assess model performance with precision, recall, and F1-score.
• Visualization: Compare model performances visually with bar graphs.

Results 📸

After training, the models' performances are evaluated and compared. The results demonstrate the effectiveness of machine learning in identifying cyberbullying content, with precision and recall metrics indicating the models' accuracy.

Summary:

• The TextCategorizer is extremely precise but lacks recall, meaning it's very conservative in its predictions. It's likely missing a lot of true cyberbullying instances.
• The TextCNNClassifier shows a more balanced and effective performance in identifying cyberbullying instances, making it a more reliable choice for practical applications.

Notes 📝

• The effectiveness of the models depends significantly on the quality and diversity of the training data.
• Further tuning and more extensive datasets might be required for optimal performance in real-world scenarios.
• This project serves as a starting point for more sophisticated cyberbullying detection systems.