Deepfake Detection on Social Media: Leveraging Deep Learning and Fast text Embeddings for Identifying Machine-Generated

Abstract

Deepfake technology, which uses AI to create manipulated media, poses a significant threat to information integrity on social media platforms. In India, the rise of deepfake content has grown exponentially, especially in the political and entertainment domains, where fake news and AI-generated videos have gone viral, leading to misinformation. The primary objective is to develop a robust AI model that accurately detects deepfake content on social media platforms, focusing on identifying machine-generated tweets using FastText embeddings. Traditional methods involved human moderation, fact-checking agencies, and manual filtering of social media posts based on predefined rules and keyword matching. These methods were time-consuming and often inaccurate, lacking the scalability to manage the massive volume of online content. The manual detection of deepfakes and AI- generated content is highly inefficient, prone to errors, and incapable of handling the vast volume of social media data in real time. As a result, harmful and misleading information can spread widely before being identified or removed. With the growing influence of social media in shaping public opinion, the motivation behind this research is to combat misinformation and safeguard the integrity of online discourse. Particularly deep learning models can significantly improve the detection of deepfakes by automating the analysis of social media content. FastText embeddings will convert tweets into meaningful word vectors, while deep learning models can be applied to classify whether a tweet is human-generated or AI-generated. This approach offers real-time detection, improved accuracy, and scalability compared to traditional methods.