Fake news detection in the Hindi language using multi‐modality via transfer and ensemble learning

Abstract Fake news classification emerged as an exciting topic for machine learning and artificial intelligence researchers. Most of the existing literature on fake news detection is based on the English language. Hence, it needs more usability. Fake news detection in low‐resource scare languages is still challenging due to the absence of large annotated datasets and tools. In this work, we propose a large‐scale Indian news dataset for the Hindi language. This dataset is constructed by scraping different reliable fact‐checking websites. The LDA approach is adopted to assign the category to news statements. Various machine‐learning and transfer learning approaches are applied to verify the authenticity of the dataset. Ensemble learning is also applied based on the low false‐positive rate of machine‐learning classifiers. A multi‐modal approach is adopted by combining LSTM with VGG‐16 and VGG‐19 classifiers. LSTM is used for textual features, while VGG‐16 and VGG‐19 are applied for image analysis. Our proposed dataset has achieved satisfactory performance.