HAN, image captioning, and forensics ensemble multimodal fake news detection
No TL;DR found
Abstract
Nowadays, news publication, propagation, and consumption have been diverted to online social media networks and web portals, which has given rise to falsified and fabricated news articles containing both textual and visual information formats. Most of the research to date is centered on textual fake news detection using machine learning approaches, where multimedia data forgery is hardly addressed. Hence, a multimodal fake news detection framework is proposed, which unitedly exploits hidden pattern extraction capabilities from text using Hierarchical Attention Network (HAN) and visual image features using image captioning and forensic analysis. We specifically focused on four different techniques of multimodal data analysis, such as HAN deep model for text, generating image caption and headline matching with news text (CHM), Noise Variance Inconsistency (NVI), and Error Level Analysis (ELA). All these algorithms have been tested, first independently and then collectively using the max voting Ensemble method on three different datasets. The experimental results and comparisons with contemporary techniques put forward the fact that the proposed method outperforms state-of-the-art with 95.90% highest accuracy on the Fake News Samples dataset. The achieved results also prove that the combined model beats individual methods' capabilities in classifying fake news accurately.