부산대학교 도서관

Currently, rumor detection models based on deep neural networks are widely used, but with the development of adversarial text technology, higher requirements are placed on the robustness of these models. In this paper, we analyze the influence of textual adversarial samples on the rumor detection model based on deep neural network for the first time. In order to accurately measure and analyze the impact of adversarial text on different structural models, we study and verify three mainstream rumor detection models on the open-source Chinese weibo rumor dataset, and analyze the readability of adversarial text and its impact on the detection performance of rumor detection model. Experimental results show that adversarial text can mislead the rumor detection model under the condition of ensuring readability, and the model using the bidirectional attention structure is more robust in the face of attacks.