Identifying Information Initiators in Mobile Social Networks: A Community-Aware Multisource Detection Framework

Abstract

The widespread adoption of mobile social networks (MSNs) has profoundly reshaped how people access and disseminate information. However, the same platforms also facilitate the rapid spread of misinformation—such as rumors, fake news, and fraudulent messages—which can lead to severe social, economic, and psychological consequences. Identifying the original initiators of such information cascades, referred to as the multiple source detection (MSD) problem, is crucial for public safety and information governance. Although prior methods have achieved promising results, most rely on specific propagation models, which restrict their applicability in real-world scenarios where the true diffusion mechanisms are often unknown or hard to estimate. Some approaches attempt to circumvent this limitation through label propagation schemes that highlight nodes surrounded by a high proportion of infected neighbors. However, the detection accuracy may be compromised due to the fact that node labels are generally represented as 1-D integers, and all infected or uninfected nodes are initialized uniformly. As a result, the structural features are not sufficiently distinguished. To address this, we propose a community-based label propagation (CLP) framework to identify numerous information sources by utilizing the community structures present in infected networks. Specifically, CLP incorporates two key effects—node prominence and exoneration—to enhance source detection. The prominence effect suggests that nodes surrounded by a higher proportion of infected neighbors are more likely to be sources. The exoneration effect relies on uninfected nodes on the boundary or infected nodes in neighboring communities to play a critical role in exonerating an infected node from being the source. Extensive experiments conducted on both synthetic networks and large-scale real-world datasets demonstrate that CLP significantly outperforms existing state-of-the-art methods across a wide range of propagation models, achieving robust, scalable, and accurate multiple source detection.