Year: 2026
Type: article
Abstract: This paper reviews recent machine learning and natural language processing work on fake news detection, with a focus on posts that combine text and images. Three recent studies form the core of the review: a systematic review of 85 papers on multilingual and low-resource detection, an explainable BiLSTM-LIME text classifier that reports 97.21% accuracy on an English news corpus, and a gated multimodal fusion network that joins IndicBERT text features with CLIP image features and reports 98.85% accuracy on a Hindi dataset. Earlier multimodal and explainability studies are used to put these results in context, and the datasets, metrics and reported figures are compared on common ground. The comparison shows that combining text with image evidence helps, that explanations are still limited to words or to a single modality weight, and that evaluation protocols differ enough that headline accuracies cannot be compared directly. Based on these findings, the paper outlines a multimodal fake news detection system for English news that analyses text and images, fuses them, explains each prediction and stores a prediction history in a web dashboard. Open challenges and future directions are also discussed
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: diamond
OpenAlex ID: W7225340708