Unifying View-Specific Learning and Ambiguity Awareness With Emotions for Multimodal Misinformation Detection
Lists
Only you see this
Abstract
Fake news detection is essential for safeguarding social media users, as the rapid spread of unverified information on these platforms facilitates the easy dissemination of fake news, creating significant risks to society. This underscores the urgent need for automated fake news detection systems that leverage advanced technologies. Current multimodal fake news detection approaches encounter challenges including inconsistent decision-making, prioritizing shared subspace information fusion over view-specific data extraction, and insufficient attention given to analyzing the sentimental elements of the news. To this end, we introduce sentiment-aware fake news detection with view-specific feature extraction (SentiView), which integrates multimodal and view-specific features and authors’ sentiments into a unified framework. The proposed framework comprises four modules: a) modal-specific encoder, which generates encodings from text and image modalities; b) multimodal consistency learning that captures the interrelationships to recognize the inherent ambiguity between the different modalities; c) view-aware feature extractor that captures information from different views within the image modality, integrating an orthogonal constraint within the shared subspace, thus facilitating the utilization of distinctive discriminative details; and d) sentiment extractor that captures the sentiments reflected to extract the underlying intention and potential bias of the news author. Extensive experiments on three publicly available datasets X, Weibo, and GossipCop show that SentiView outperforms state-of-the-art fake news detection approaches with an accuracy of 93.8%, 94.5%, and 87.3%, on the datasets respectively.
BibTeX
@article{malik2026unifying,
author = {Marium Malik and Yang Song and Jiaojiao Jiang and Sanjay Jha},
title = {{Unifying View-Specific Learning and Ambiguity Awareness With Emotions for Multimodal Misinformation Detection}},
journal = {IEEE Trans. Comput. Soc. Syst.},
year = {2026},
doi = {10.1109/tcss.2025.3626946},
}Loading related content...