MDN：Meta Detection Network for Fake News Detection
The MDN algorithm uses a model-agnostic meta-learning algorithm to divide the model’s training into event adaptation and specific event detection stages, enabling the model to learn from existing knowledge and adjust parameters based on new tasks.
Model Architecture Diagram：
Description： The MDN method has been proposed to address the issue of poor generalization performance of models due to the difficulty of collecting high-quality relevant annotation data for new events in a short time. It utilizes a model-agnostic meta-learning algorithm (MAML) to divide the model’s training into an event-adaptive stage and a specific event detection stage, allowing the model to leverage existing knowledge for learning and then adjust parameters based on the new task. A multi-modal feature extractor is also constructed to combine high-dimensional features of tweet texts and images.