Fake News Classification Using Bilstm Deep Learning Algorithm

Abstract

In recent years, the rapid dissemination of information has been paralleled by the equally swift spread of fake news, making it increasingly challenging to discern truth from falsehood. Fake news has had detrimental effects on politics and culture. This research presents the design and methodology for detecting fake news, with a particular focus on a deep learning model employing a Bidirectional Long Short-Term Memory (BiLSTM) network enhanced by an attention mechanism. The BiLSTM network captures contextual information from both past and future states of a sequence, while the attention mechanism emphasizes the most relevant input data , thereby improving the model’s ability to focus on critical information. These combined features enhance the accuracy and reliability of fake news detection. The model is evaluated using Fakeddit dataset, which comprises of news articles sourced from Reddit website. The models obtained an accuracy of 73% for both the BiLSTM and attention BiLSTM. Model created offers effective tools for combating fake news. Attention mechanism was not able to achieve its objective and hence need for use in different datasets or application of optimization techniques. Future research should be able focus on enhancing data diversity, model efficiency, and the use of the model in various languages in the world.