Multimodal social media disinformation detection by fusing images and BERT

To address the diversity and fragmentation issues in false information detection (FID) on social media, this article proposed the MM-GCN-BERT model based on the advantages of GCN and BERT models, integrating multimodal fusion and cascading detection frameworks. By designing and optimising the MM-GCN-BERT model, and combining specific experimental data to verify the performance of the MM-GCN-BERT model, experimental results were obtained. The results showed that the average F1 value of the MM-GCN-BERT model was 0.816, the average precision was 0.836, and the average recall was 0.797. The indicators of the model are all at a high level. It can be seen that the MM-GCN-BERT model has shown excellent performance in solving the problems of diversity and fragmentation in detecting false information on social media.