Malay Fake News Classification Using a Deep Learning Approach

Abstract

In today’s world, most people spend their time interacting online through social media platforms. They regularly consume news from online media sources such as social media and online newspapers. People also tend to spread the information received without confirming its truth. As the spread of fake news online grows, the public and researchers have become more interested in identifying fake news. The content of fake news that has manipulative information has negative effects on individuals and society. Most existing research investigates different kinds of features from news articles to develop fake news detection. The existing research is not capable of determining fake news for the Malay language due to lake of resources for Malay Language. The goal of this paper is to evaluate the performance of the proposed Bidirectional RNN deep learning approach to classify fake Malay news by varying the dropout rate of the proposed RNN model. In this work, four different dropout’s values are used to evaluate the performance of the RNN models, which are 0.1 (Model M1), 0.3 (Model M2), 0.5 (Model M3) and 0.8 (Model M4). Based on the results obtained, the smaller is the dropout rate the less number of epochs is required to train the RNN model as it is observed that Model M1 required more number of epochs to train the RNN model. However, the best accuracy can be obtained when the dropout rate was 0.3 with 90.1% accuracy in Model M2. Having higher percentage of dropout rate will not produce a trained model that can perform with high accuracy value. In conclusion, maintaining the percentage of dropout to be 0.3 and below enables the LSTM to produce good values of accuracies. The accuracy of the forecasted result is highly influenced by the length of the text.