Twitter Bot Detection using Deep Learning

This repository contains code and resources for detecting Twitter bots using deep learning techniques. The project aims to identify and classify Twitter accounts and tweets as bots or non-bots based on various features extracted from their profiles and activities.

Table of Contents

• Introduction
• Dataset
• Model Architecture
• Usage
• Evaluation
• Results
• Contributors

Introduction

With the increasing presence of bots on social media platforms, it has become important to develop effective methods for detecting and identifying these automated accounts. This project focuses on Twitter bot detection using a deep learning approach. The model is trained on a dataset of bot and non-bot Twitter accounts and tweets, and it learns to classify new accounts based on their features.

Dataset

The dataset used for training and evaluation is cresci-2017, a combination of bot and non-bot Twitter accounts. It includes features such as username, age, location, verification status, tweet count, follower count, and more. The dataset is divided into training and test sets to train the model and evaluate its performance. Publicily available at here

Model Architecture

The model architecture used for Twitter bot detection is a deep neural network. It consists of multiple dense layers with ReLU activation and a final sigmoid activation layer for binary classification. The model takes various account features as input and predicts the probability of an account being a bot.

Usage

1. Clone the repository:

   git clone https://github.com/your-username/twitter-bot-detection.git

2. Install the required dependencies:

   pip install -r requirements.txt

3. Run the Flask app:

   python app.py

4. Access the web application in your browser at http://localhost:5000/home.

Evaluation

The model is evaluated on the test dataset using metrics such as precision, recall, F1 score, and ROC/AUC score. These metrics provide an assessment of the model's performance in correctly classifying bot and non-bot accounts. Flask app is built to test for real world data.

Results

Account-Level Classification

The model achieves high precision, recall, F1 score, and ROC/AUC score, indicating its effectiveness in detecting Twitter bots. However, it's important to note that the performance may vary depending on the dataset and real-world scenarios.

Metric	Score
Recall Score	0.9756345177664975
F1 Score	0.9821154828819622
ROC/AUC Score	0.9962218266786795

Tweet-Level Classification

The model achieves good precision, recall, F1 score, and ROC/AUC score, indicating its effectiveness in detecting Twitter bots. However, it's important to note that the performance may vary depending on the dataset and real-world scenarios.

Metric	Score
Recall Score	0.8872471207100318
F1 Score	0.834560110730113
ROC/AUC Score	0.8468545091803743

Contributors

• Muhammad Hassaan Ibrahim
• Ahmad Faraz Sheikh