DAFD - Domain Adaptation Framework for Fake News Detection.

Nowadays, social media has become the leading platform for news dissemination and consumption. Due to the convenience of social media platforms, fake news spread at an unprecedented speed, which has brought severe adverse effects to society. In recent years, the method based on deep learning has shown superior performance in fake news detection. However, the training of this kind of model needs a large amount of labeled data. When a new domain of fake news appears, it usually contains only a small amount of labeled data. We proposed a novel Domain Adaptation framework for Fake news Detection named DAFD. It adopts a dual strategy based on domain adaptation and adversarial training, aligns the data distribution of the source domain and target domain during the pre-training process, and generates adversarial examples in the embedding space during the fine-tuning process to increase the generalization and robustness of the model, which can effectively detect fake news in a new domain. Extensive experiments on real datasets show that the proposed DAFD achieves the best performance compared with the state-of-the-art methods for a new domain with a small amount of labeled data.