TPSD: A Two-Phase Framework for Rumor Source Identification in Social Networks

The rapid spread of rumors in online social networks creates serious challenges for information credibility and public trust. Identifying the origin of rumor propagation is therefore essential for controlling misinformation and minimizing its harmful impact. Existing rumor source detection methods often suffer from high computational complexity and reduced accuracy in large-scale networks. This paper proposes a Two-Phase Source Detection (TPSD) framework for efficient rumor source identification in social networks. The proposed method first minimizes the search space by partitioning the infected graph into nominee communities using diffusion characteristics and infection timestamps. In the second phase, observer nodes with high betweenness centrality are selected to estimate the actual rumor source using reverse propagation. The Susceptible-Infected (SI) diffusion model is employed to simulate rumor dissemination on synthetic Erdős–Rényi (ER) graphs and real-world Facebook and Twitter datasets. The performance of the proposed method is evaluated using Distance Error (DE) and Average Distance Error (ADE). Experimental results demonstrate that the proposed TPSD approach achieves superior performance compared to existing methods such as PTVA and Louni. The proposed model reduces the source detection error from 0–6 hops to 0–1 hops on benchmark datasets while significantly decreasing computational cost through network reduction. The results confirm that TPSD provides an accurate and computationally efficient solution for rumor source identification in social networks.