A Hybrid CNN-Based Approach for Detecting Machine-Generated Tweets Using Fast Text Embeddings

Abstract

The rapid growth of social media platforms has led to an increase in the spread of misleading and automated content, particularly through deep fake tweets generated by bots. This project focuses on detecting such fake tweets using machine learning and deep learning techniques. The system allows users to load datasets, convert textual data into numerical vectors using FastText embedding, and train multiple algorithms to evaluate performance. Among the applied models, Convolutional Neural Network (CNN) and hybrid CNN approaches demonstrate superior accuracy in identifying fake content. The application provides a user-friendly interface for login, dataset loading, model training, and real-time prediction of tweets. By analyzing textual patterns and semantic features, the system effectively distinguishes between human-written and bot-generated tweets. This approach contributes to enhancing trust and reliability in social media communication by reducing the impact of automated misinformation.