Analyzing Cyberbullying Tweets with LSTM Networks

This project aims to develop a tool for identifying cyberbullying tweets and classifying them based on various categories such as gender, religion, age, ethnicity, and other types of cyberbullying. The primary objectives include:

• Utilizing the Cyberbullying Classification Dataset sourced from Kaggle.
• Conducting data cleaning procedures to enhance data quality.
• Applying data preprocessing techniques to prepare the cleaned data for analysis.
• Constructing a Recurrent Neural Network (RNN) model using Long Short-Term Memory (LSTM) layers and evaluating its performance on a separate test dataset.
• Implementing a client-facing API using Flask for seamless integration and usability.

Technologies and Resources

• Python Version: 3.10

• Libraries: numpy, pandas, matplotlib, seaborn, nltk, tensorflow, scikit-learn, flask, json

• Flask API Setup:

  • pip install -r requirements.txt
  • conda env create -n <ENVNAME> -f environment.yaml (Anaconda environment)

• Dataset: https://www.kaggle.com/datasets/andrewmvd/cyberbullying-classification

Data Acquisition

The project relies on the Cyberbullying Classification Dataset obtained from Kaggle. This dataset comprises over 47,000 labeled tweets categorized into distinct classes of cyberbullying.

• Not Cyberbullying
• Gender
• Religion
• Other types of cyberbullying
• Age
• Ethnicity

Data Cleaning

A custom Python script is developed to perform rigorous data cleaning processes. These processes involve:

• Removal of punctuation marks
• Elimination of numerical characters
• Conversion of text to lowercase
• Elimination of stop words
• Lemmatization/Stemming of words
• Removal of URLs

Data Preprocessing

To prepare the cleaned tweets for analysis, the TextVectorization layer from Keras is applied. This layer facilitates one-hot encoding of text, resulting in a list of encoded integers representing individual words (or tokens) in the input string. Additionally, sequences are padded to ensure uniform length.

Model Building

1. Train-Test Split: Data is divided into 80% training and 20% testing sets.
2. Bidirectional LSTM Model: Build an RNN architecture utilizing Bidirectional LSTM layers.
3. Evaluation: Employ "categorical_crossentropy" for loss measurement and "RMSprop" for optimization.

Model Visualization:

Model Performance:

Productionization

A Flask-based user interface (UI) allows users to submit tweets and receive cyberbullying type predictions in real-time.