Addressing misinformation in Bengali media: A hybrid deep learning solution

Abstract

The spread of fake news poses significant threats to society, democracy, and public discourse. This is especially true in Bengali-speaking communities where digital literacy and language-specific resources are limited. To address this issue, we urgently need effective ways to detect and mitigate fake news in the Bengali language. This paper highlights the unique challenges associated with linguistic diversity and misinformation prevalence. Our research introduces a hybrid deep learning algorithm specifically designed to classify fake and authentic Bengali news articles. The methodology involves extensive data collection, rigorous preprocessing to improve textual quality, and the use of advanced feature extraction methods. We combined and trained CNN and LSTM/BiLSTM models to handle the nuances of Bengali text. Our proposed algorithm significantly outperforms existing models, with a high accuracy of 98.45% in distinguishing between fake and authentic news and reducing false positives. Additionally, we developed a user-friendly web application that allows general public to input news articles and obtain predictions on their authenticity in real-time. This study not only advances the field of fake news detection but also provides essential tools for journalists, policymakers, and the general public to combat misinformation in Bengali media. Our findings have broader implications, promoting informed public discourse and supporting the development of resilient digital ecosystems. This underscores the need for ongoing research and adaptation of such models to combat the global challenge of fake news in various languages and contexts.