Opinions on social media can be swayed thanks to new developments in natural language crea-tion. The capacity of deep neural networks to generate content has also been enhanced via lan-guage modeling. Because of this, text-generative algorithms have improved to the point that at-tackers may train social bots to publish deepfakes that seem legitimate and sway public opinion. Reliable and precise deepfake social media message detection systems are required to address this issue. Keeping this in mind, the present research finds Twitter posts that are made by machines. Using a basic deep learning model and word embeddings, this study leverages the publicly avail-able Tweepfake dataset to distinguish between human and bot-generated tweets. A standard con-volutional neural network (CNN) architecture is trained to detect deepfake tweets using FastText word embeddings. In order to prove that the suggested strategy is better, this research compared it to various machine learning models that served as baselines. Here are some baseline approaches: FastText, Term Frequency, FastText subword embeddings, and Term Frequency Inverse Docu-ment Frequency. We also compare the proposed method to other deep learning models, such as CNN-LSTM and Long short-term memory (LSTM), to demonstrate its efficacy and utility in solving the problem. The CNN architecture, when combined with FastText embeddings, efficient-ly and correctly classifies twitter data with a 93% accuracy rate, according to the experimental results.
Reddy et al. (Thu,) studied this question.