Feature fusion for multimodal fake news detection

With the rise of social media, people increasingly rely on these platforms to stay informed about global events. However, the credibility of these sources remains uncertain. The accessibility and speed that make social media appealing have also contributed to a growing issue of fake news. But, the very nature of the social media where due to the lack of fact check on the news article along with fabricated visuals, creates a sense of confusion among people. To address this problem, this paper presents feature fusion techniques for multimodal fake news detection problem. The framework uses the VGG-19 pre-trained model to preprocess and extract visual features, and GloVE is used for generating text embeddings, these features are fused using three different fusion techniques (1) concatenation, (2) addition and (3) weighted addition. This paper presents a comparative analysis of these feature fusion techniques on FakeNewsNet dataset. The results shows the supremacy of weighted addition algorithm over the other two algorithms over various performance metrics.