BRaG: a hybrid multi-feature framework for fake news detection on social media

Social media has gained immense popularity for its convenience, affordability, and interactive features. However, the characteristics that make social media platforms appealing also provide a fertile ground for the spread of fake news–deliberately misleading and unverifiable information that can have severe consequences for individuals and society. Previous approaches for detecting fake news have mostly focused on single aspects such as text, but are inadequate as fake news evolves to closely resemble genuine news. To enhance fake news detection, a comprehensive multi-faceted approach is necessary. Various machine-learning techniques have been used to detect fake news. This paper introduces a novel hybrid and multi-feature framework for detecting fake news that considers both the content (e.g., text) and context (e.g., user profiles and propagation graph) of news. Our framework, BRaG, leverages a combination of the BERT pre-trained language model, recurrent neural network (RNN), and graph neural network (GNN) to analyze news text, sequence of engaged users, and the estimated news propagation graph, respectively, and form the final news representation vector. Additionally, our approach incorporates text emoji meanings to take into account the contextual information they convey. The proposed framework is evaluated on two real-world datasets and outperforms existing baselines and state-of-the-art fake news detection models.