Abstract :

Recent advancements in natural language processing and generative language modeling have significantly enhanced the capabilities of deep neural networks in content creation. While these developments offer numerous benefits, they also raise serious concerns regarding the misuse of text-generative models to produce deepfake content on social media platforms. Such content, often disseminated by sophisticated social bots, can manipulate public opinion and spread misinformation. This research focuses on detecting machine-generated content on Twitter using the publicly available Tweepfake dataset. A deep learning approach based on a Convolutional Neural Network (CNN) architecture, integrated with FastText word embeddings, is proposed to classify tweets as either human-generated or bot-generated. To evaluate the effectiveness of the proposed model, it is compared against several baseline machine learning methods, including models using Term Frequency-Inverse Document Frequency (TF-IDF), FastT