FAST TEXT EMBEDDINGS AND DEEP LEARNING FOR ROBUST DETECTION OF DEEPFAKES IN SOCIAL MEDIA TWEETS

Abstract

The increasing prevalence of deep fake technology has prompted apprehension regarding the dissemination of inaccurate information on social media. This paper illustrates a deep learning-based approach to identifying deep fake tweets, particularly those generated by machines. This will mitigate the detrimental effects of false information on the internet. Our approach categorizes tweets into categories by employing Fast Text embeddings and deep learning models. We employ Fast Text embeddings to generate dense vector models after preprocessing the tweet text. The distinction between genuine and fraudulent tweets is determined by the semantic information regarding tweet topics that these embeddings accumulate. We incorporate these embeddings into a deep learning model, such as a CNN or a Long Short-Term Memory (LSTM) network, to determine whether the tweets are genuine or fabricated. Machine-generated tweets are generated using contemporary text generation algorithms that have been instructed on a collection of tagged tweets. Research conducted on a real-world tweet collection demonstrates that our methodology is effective in identifying tweets that were generated by algorithms. Our approach is significantly more precise than other methods for identifying social media deep fakes. In general, our proposed approach is a dependable and efficient approach to identify tweets that were generated by machines and to halt the dissemination of inaccurate information on social media.