Recognizing fake news spreading in social networks using ensemble learning

Fake news on social media is rapidly spreading, causing significant damage and highlighting the need for effective communication strategies to detect and combat its spread. This study introduces a method that utilizes a combination of Deep Learning (DL) techniques for identifying fake news-related messages on social networks. Each deep model receives unique representations of the message’s content features, and the process begins with preprocessing the content to obtain a standardized intermediate form. Three DL models are employed: the first is a multilayer perceptron with two hidden layers, which processes statistical features of the message and detects fakeness based on content properties and dissemination statistics. The second and third models are convolutional neural networks (CNNs) trained using TF-IDF and Word2Vec attributes, considering the content features of the message in detecting fake news. After training these models, they are then used for recognizing fake news in new samples. The presented method used a decision tree-based learning model in the final step for result ensemble. In comparison to conventional ensemble systems like majority voting, this approach significantly increases the flexibility of the model. In our experiments, we utilized two databases: GossipCop and PolitiFact. Our proposed method performed well in both databases, achieving an average accuracy of 99% and 96% respectively. This demonstrates that our approach outperforms comparative methods in detection.