Securing Consumer Applications Against AI-Driven Misinformation: A Cross-Domain Multimodal Approach

Consumer platforms such as smart healthcare and digital wellness, smart education and online learning, and smart technology and consumer electronics are increasingly exposed to AI-driven misinformation. Typical attacks include machine-generated text that is misleading, manipulated images, and multimodal posts in which text and images are semantically inconsistent, which can erode user trust and disrupt critical decision-making. Most existing detection methods are developed for a single domain or modality and suffer from severe performance degradation under cross-domain distribution shifts and text-image inconsistencies. To address these challenges, this article proposes a cross-domain multimodal misinformation detection framework (CDMMD) for text–image posts on consumer platforms. CDMMD performs multi-perspective feature extraction and employs a domain-aware dual-expert network (D2EN) to refine shared and domain-specific knowledge. A partially parameter-shared recursive cross-modal attention mechanism is then used to progressively fuse global semantics with heterogeneous text–image features, and a lightweight MLP-induced adaptive convolution module reweights the fused knowledge to enhance robustness. Experiments on two Weibo-based benchmark datasets and the FineFake dataset, covering multiple consumer scenarios such as healthcare, education, and technology, demonstrate that CDMMD consistently outperforms strong baselines in cross-domain multimodal detection. Across these datasets, CDMMD improves accuracy by up to 1.8% points over the best competing methods, indicating its effectiveness in improving content credibility assessment on consumer platforms and providing reliable signals for downstream tasks such as propagation modeling and attacker tracking.