Prathima Patnaik,M.koushik Reddy, K.Bala Krishna 02 10 The rapid advancements in natural language generation have introduced powerful tools capable of shaping public opinion on social media platforms Enhanced language modeling techniques have significantly improved the generative capabilities of deep neural networks enabling them to produce highly realistic and contextually accurate text This progress has led to the emergence of sophisticated textgenerative models that can be exploited by adversaries to power social bots creating convincing deepfake posts that influence public discourse To combat this growing threat the development of robust and accurate detection methods for identifying machinegenerated content is essential In response this study focuses on the detection of deepfake tweets on platforms such as Twitter A deep learningbased approach is proposed employing a Convolutional Neural Network CNN architecture in combination with FastText word embeddings to classify tweets as either humangenerated or botgenerated The model is trained and evaluated using the publicly available Tweepfake dataset To validate the effectiveness of the proposed method it is benchmarked against several traditional machine learning models utilizing features like Term Frequency TF Term Frequency Inverse Document Frequency TFIDF FastText and FastText subword embeddings Additionally comparisons are made with other deep learning architectures including Long ShortTerm Memory LSTM and hybrid CNNLSTM models Copyright (c) Journal of Science Engineering Technology and Management Science. All rights reserved 2026