Recent multimodal frameworks often grapple with semantic misalignment and noise, impeding effective integration of diverse modalities. In order to solve this problem, this study presents CaMN (Cross-aligned Multimodal Network), a framework designed to enhance multimodal understanding through a robust cross-alignment mechanism. Unlike conventional fusion methods, our framework aligns features extracted from images, text, and graphs via a tailored loss function, enabling seamless integration and exploitation of complementary information. Leveraging Abstract Meaning Representation (AMR), we extract intricate semantic structures from textual data, enriching the multi-modal representation with contextual depth. Furthermore, to enhance robustness, we employ a masked autoencoder to simulate noise-independent feature space. Through comprehensive evaluation on the crisisMMD dataset, CaMN demonstrates superior performance in crisis event classification tasks, highlighting its potential in advancing multimodal understanding across diverse domains. Our code is available at https://github.com/brillard1/CaMN.

Proliferation of fake news has become a critical issue in today's information-driven society. Our study includes external knowledge from Wikidata which allows the model to cross-reference factual claims with established knowledge. This approach deviates from the reliance on social information to detect fake news that many state-of-the-art (SOTA) fact-checking models adopt. This paper introducesEA$^{2}$2N, anEvidence-basedAMR (abstract meaning representation)AttentionNetwork for Fake News Detection. EA$^{2}$N utilizes the proposed Evidence based Abstract Meaning Representation (WikiAMR) which incorporates knowledge using a proposed evidence-linking algorithm, pushing the boundaries of fake news detection. The proposed framework encompasses a combination of a novel language encoder and a graph encoder to detect fake news. While the language encoder effectively combines transformer-encoded textual features with affective lexical features, the graph encoder encodes semantic relations with evidence through external knowledge, referred to as WikiAMR graph. A path-aware graph learning module is designed to capture crucial semantic relationships among entities over evidence. Extensive experiments support our model's superior performance, surpassing SOTA methodologies with a difference of 2-3% in F1-score and accuracy for Politifact and Gossipcop datasets. The improvement due to the introduction of WikiAMR is found to be statistically significant with t-value less than 0.01.