Distinguishing Abstracts of Human-Written and ChatGPT-Generated Papers in the Field of Computer Science

Abstract

In the era of artificial intelligence, advanced technologies such as text deep fakes have emerged, utilizing AI and deep learning algorithms to generate text contents that convincingly mimic reality. Text-based deep fakes, commonly found in emails, articles, news, and social media posts, pose a significant threat to public trust. Our research proposes a method to distinguish between fake and genuine scientific abstracts in the field of computer science using a specialized dataset and deep learning models. The proposed detection model can differentiate between real and fake abstracts with 97.5% accuracy. Additionally, we employed metrics such as precision, recall, accuracy, and F1 score to measure the performance of our system in detail.