ارائه چارچوبي براي تشخيص زودهنگام شايعات فارسي مبتني بر شبكه مولد متخاصم

Social media platforms play a significant role in connecting individuals, disseminating information, and spreading messages, becoming one of the primary sources of news for people worldwide. Studies show that vast and extensive amounts of data are exchanged across these platforms among different communities and individuals, making the verification and validation of such information essential. Particularly in recent years, social media has been increasingly used as a medium for spreading rumors. These rumors often cover a variety of areas, including political, social, cultural, and tourism topics, and they can inflict irreparable harm on individuals and society. Therefore, it is imperative to react pro‎mp‎tly to rumors and take measures to prevent their spread. Various methods and approaches have been developed to detect and curb the spread of rumors, many of which are based on artificial intelligence and machine learning. In this study, using a Generative Adversarial Network (GAN) approach and leveraging large language models (LLMs), a framework is proposed to detect rumors in their early stages by augmenting data and generating synthetic texts similar to the original samples. The primary innovation of this research lies in early detection of rumors in the Persian language, as the framework is designed to focus specifically on identifying rumors within the initial hours of their dissemination. To the best of our knowledge, no prior research has been conducted in this domain for the Persian language. In this study, the proposed model achieves an accuracy of 83% and an F1-score of approximately 88% in identifying rumors, preventing their widespread propagation. This research represents a step forward in combating rumor-spreading phenomena in Persian and can significantly contribute to improving information quality and raising public awareness. Furthermore, the results of this model have been compared with traditional methods, and the findings demonstrate that despite being trained on a limited set of rumors, the model performs effectively in this area.

6