Spreading Mosaic: An Image Restoration-Inspired Social Rumor Propagation Model

This article draws inspiration from deep learning image restoration technology. If users involved in the rumor topic are regarded as pixels in an image, the uncertainty of user behavior is similar to the ambiguity of pixels in a mosaic image. The prediction of user behavior is influenced by the user and neighboring friends. Similarly, the recovery of mosaic image pixels is also influenced by these pixels and neighboring pixels. Thus, during rumor propagation, the prediction of user behavior is equivalent to the restoration of pixels in the mosaic image. Based on this inspiration, this study proposes a rumor propagation prediction model based on image restoration technology. First, we propose the concept of topic images and design the rumor2pixel algorithm to pixelate the topic of rumor propagation. Second, through the Generative Adversarial Network model, fuzzy pixels in the “rumor topic image” are compensated to learn more realistic rumor propagation trends. Finally, a dynamic approach for predicting the propagation of rumor and countering it based on evolutionary game theory is proposed, named Rumor-DPM (rumor dynamic propagation model). This approach is focused on reconstructing rumor images while taking into account the conflict between rumors and anti-rumors as well as its timeliness. The experimental findings demonstrate that this strategy can more accurately depict the internal dynamics between rumors and anti-rumors and effectively and successfully improve the ability to forecast user behavior throughout the rumor-propagation process.