A novel attention-based multimodal framework for automatically detecting cryptocurrency fake news

Abstract

Cryptocurrencies are increasingly the subject of fake news, increasing risks for market stability and investor decisions. To address this issue, we propose a multimodal framework to detect fake cryptocurrency news using text, image, and sentiment features with BERT, Swin Transformer, and RoBERTa, respectively. We use multi-head attention to combine these features to ensure the complementarity of features from different modalities. The fused representations are passed into a fully connected layer for final classification. Experimental results show that this framework achieves better accuracy and reliability than unimodal and multimodal models for detecting cryptocurrency misinformation.