Fake news detection based on multi-feature fusion under attention guidance

Convolutional Neural Networks (CNNs) can only extract local text features and lack the representation of long-term text features, whilst Recurrent Neural Network (RNN) suffers from the gradient vanishing problem, which affects the text classification result. Aiming at these problems, we develop a fake news detection model based on multi-feature fusion under attention guidance. Firstly, we obtain the word vectors by the Global Vectors for Word Representation (GloVe) model. Secondly, we establish the fake news detection model. Local and deep semantic features of texts are extracted by CNN and Bidirectional Gated Recurrent Unit (BiGRU), respectively. The key features are examined by the attention mechanism. Finally, local features and deep semantic features are combined to achieve text classification. Experiments show that the proposed model's text classification accuracy reaches 95.74%.