With the rapid development of the Internet and social media, the spread of fake news has become a global problem, posing a serious threat to the authenticity of information and social stability. Consequently, to enhance the efficacy of multimodal information integration and augment the robustness and detection accuracy of the model in complex fake news scenarios, a set of approaches is proposed. These approaches include fake news detection models based on visual features and enhanced generative adversarial networks. Multimodal information such as text, image and video are extracted by feature extractors to improve the generalization and robustness of the model. Test results showed that the optimal parameters of the visual feature model were 70 learners, 0.01 learning rate, and a maximum tree depth of 3. Its accuracy, recall, recall, and F1-scores in the two datasets were 90.23%, 89.44%, 88.95%, and 0.89, respectively. The accuracy, recall, recall, F1, and area under the curve of the generative adversarial network model were 93.42%, 92.51%, 92.61%, 0.92, and 0.93, respectively. The experimental findings demonstrate that the two proposed detection models exhibit superiority over traditional multimodal fusion methods in terms of accuracy and robustness. The study demonstrates that by effectively integrating multimodal information and adversarial training mechanisms, the overall performance of fake news detection systems can be significantly enhanced, offering novel approaches for addressing complex detection tasks.
Zheng Ji (Sun,) studied this question.