A Hybrid Deep Learning Architecture for Misinformation Detection on Social Media

Social media has grown to become a popular source of news and information for users around the world. However, the strength of fast dissemination of information to users in diverse places also exposes social media platforms, including Twitter, to the spread of misinformation, such as rumors or false information. Automated classification of such news on social media is a challenging task. In this study, we propose a hybrid deep learning model that utilizes a Features-Based model (FB) extracted from two levels: tweet level and user level, combined with pre-trained text embedding models such as Global Vectors for word representation (GloVe) and Universal Sentence Encoders (USE). The models were evaluated on a real-world dataset containing a collection of Twitter rumors and non-rumors. The experimental evaluation results reveal that our hybrid deep-learning model achieves higher accuracy in detecting rumors compared to the baseline learners and previous methods. Further, a hybrid model that combined a features-based model and text embedding model led to improve performance compared to use a single model.