ABSTRACT The widespread rise of impersonation, coordinated inauthentic behavior, and AI‐generated identities on social media has created an urgent need for robust and trustworthy profile verification systems. Existing multimodal approaches such as various Convolutional Neural Network (CNN) architectures, traditional Long Short‐Term Memory (LSTM)/Gated Recurrent Unit (GRU) models demonstrate progress but remain limited by weak modality synchronization, insufficient modeling of user–content relationships, and lack of graph‐structured reasoning necessary to detect sophisticated fake networks. This study addresses these gaps by introducing a unified multi‐modal and heterogeneous graph reasoning framework designed to authenticate social media profiles using text, images, and behavioral metadata. The proposed method synchronizes modalities via temporal alignment and missing‐data reconstruction, constructs a heterogeneous user–content–interaction graph with typed nodes, weighted edges, and contextual attributes, and employs Graph Neural Networks (GNNs) (Heterogeneous Graph Transformer (HGT), Graph Attention Network (GAT), Heterogeneous Graph Attention Network (HAN)) combined with an attention‐based late fusion mechanism to integrate relational and modality‐specific features. A curated multi‐platform dataset containing 52,000 user profiles, posts, and interaction histories was developed to evaluate system performance. The proposed framework achieved strong quantitative results on the multimodal dataset, attaining 94.7% accuracy, 93.4% precision, 95.8% recall, and an F1‐score of 94.6%, demonstrating robust discriminatory capability across diverse profile types. Experimental results show consistent improvements of 6%–12% in F1‐score and accuracy, with statistically significant performance gains ( p < 0.05). The study concludes that combining multi‐modal deep learning with heterogeneous graph reasoning substantially enhances authenticity verification and improves resilience to evolving deceptive behaviors.
Rami Baazeem (Mon,) studied this question.