The rapid growth of social media platforms has significantly increased the spread of fake news, posing serious threats to public trust, democratic processes and social harmony. Traditional fake news detection methods relying solely on textual analysis are often insufficient as modern misinformation is increasingly multimodal, combining text, images, videos and metadata to enhance credibility and emotional impact. This thesis proposes a multimodal machine learning framework for the early detection of fake news by jointly analysing textual, visual and contextual features. By integrating natural language processing techniques with deep visual feature extraction and multimodal fusion strategies the proposed approach aims to improve detection accuracy at early stages of news dissemination. Experimental evaluation on benchmark datasets demonstrates that multimodal models outperform unimodal approaches highlighting the importance of cross modal correlations in identifying deceptive content.
Bara et al. (Thu,) studied this question.