Early Rumor Detection based on Data Augmentation and Pre-training Transformer

The development of online social networks such as Sina Weibo, Twitter and Facebook has completely changed the way people communicate. While online social media makes it easier for people to get information, it also results in proliferation and wide spread of rumors. Automatic detection of rumors has become a hot topic of research. However, detecting rumors in the early stage is quite challenging. This paper proposes a new model (DAPT) for early rumor detection by analyzing text features, using pre-training technique, and employing data augmentation method. Experiments on BuzzFeed, PolitiFact, and Twitter 16 datasets show that the accuracy and F1-score by the proposed model are better than the existing models.