Enhancing Bangla Fake News Detection By Using A Deep Learning Approach

The spread of fake news, particularly in the Bangla language, seriously jeopardizes media literacy and social harmony. This research proposes a new approach in fake news detection in Bangla using the neural network-artificial intelligence technique Long Short-Term Memory. In this regard, a dataset of news articles totalling 42500 articles was gathered from various online platforms; the dataset had a heavy imbalance between fake and real news distribution, so such imbalance had been treated using oversampling techniques. It is tuned with the best architecture, having an embedding layer and LSTM layer with fully connected layers at the top for binary classification. The model achieved 95% accuracy in the classifications. This, therefore, confirms further the robustness of the model in detecting fake news in the Bangla language using precision, recall, and F1-score evaluation metrics. The present study explores LSTMs as variants of deep learning models very effective at handling sequential data in low-resource languages for tackling misinformation.