An Incomplete Observation-Oriented Approach to Rumor Source Identification

An efficient SI model-based approach, HiSOR, is proposed to address the insufficient localization accuracy and elevated computational complexity of timestamp-based algorithms for identifying rumor sources in sparse observation environments. In HiSOR, a local structural saliency score is calculated for each node within its multi-hop neighborhood in the infected subgraph. A certain percentage of nodes with the highest saliency scores is subsequently chosen to form the observer set. Two principal observers act as anchors to construct a crucial region and radial propagation trees, while a scoring function for propagation deviation is utilized to rate potential source nodes. The node with the lowest score is finally returned as the estimated source. Comprehensive experiments were performed, revealing that HiSOR, in comparison to current localization techniques, attains reduced distance errors, enhanced identification accuracy, and improved runtime efficiency, thus providing a precise and effective solution for rumor tracing in social networks.