Using deep learning techniques to identify false news in Bangla

Abstract

The prevalence of fake news has increased the need for sophisticated detection algorithms, particularly those that operate in languages other than English. The purpose of the study is to address the issues surrounding Bangla, an indigenous tongue that is viewed as less important. It is advised to use a full data with approximately 50,000 articles to do this. Many deep learning models have been tested with this dataset, including hybrid architectures, reversible gated recurrent units (GRU), long short-term memory (LSTM), and 1D convolutional neural network models (CNNs). Several useful criteria, including retention, accuracy, F1 score, and efficiency, were used in this study to assess the model's efficacy. A sizable software was used to do this. By using the Bidirectional GRU models, we are able to achieve an amazing 99.14% accuracy rate in identifying erroneous information in Bangla. We carry out comprehensive tests to prove the effectiveness of these models in this regard. The significance of preserving dataset balance and the amount of continuous improvement work needed are highlighted by our research. By employing a prototype online application and a web application machine learning model classifier to identify fake news in Bangla, this study establishes the foundation for future developments in the detection process. In particular, it makes a major and low-resource contribution to the advancement of Bangla false news detection systems.