A Deep Learning Ensemble Framework for Fake News Stance Detection and Classification

Abstract

The method of stance identification in misleading information is crucial to determining the credibility of the news since it aids in fact-checking by identifying diverse sources' positions on a primary assertion. To address the issues that arose throughout the process of learning ML models, several research projects included ensemble learning into ML models. The fundamental difficulty with models based on deep learning is that it takes a lot of skill and expertise to fine-tune the ideal hyperparameters in order to achieve a global minimal error. However, identifying the best hyperparameters necessitates a time-consuming approach in the field of search, which makes the work tiresome. The development of Ensembles Deep Learning Models (EDLM) for identifying and categorizing bogus news into predefined fine-grained categories is the focus of this work. Convolutional neural networks (CNN), LSTM (Long Term Memory), as well as bi-directional long short-term memory (Bi-LSTM) networks should be used as the foundation for ensemble models at initially. For the final classification, the representations produced by the two aforementioned algorithms are incorporated through a perceptron with multiple layers model (MLP). Experimental analysis demonstrates that our suggested ensemble learner technique outperforms individual learners.