Tracking Clustering Coefficient on Dynamic Graph via Incremental Random Walk

Abstract

Clustering coefficient is an important measure in complex graph analysis. Tracking clustering coefficient on dynamic graphs, such as Web, social networks and mobile networks, can help in spam detection, community mining and many other applications. However, it is expensive to compute clustering coefficient for real-world graphs, especially for large and evolving graphs. Aiming to track the clustering coefficient on dynamic graph efficiently, we propose an incremental algorithm. It estimates the average and global clustering coefficient via random walk and stores the random walk path. As the graph evolves, the proposed algorithm reconstructs the stored random walk path and updates the estimates incrementally. Theoretical analysis indicates that the proposed algorithm is practical and efficient. Extensive experiments on real-world graphs also demonstrate that the proposed algorithm performs as well as a state-of-art random walk based algorithm in accuracy and reduces the running time of tracking the clustering coefficient on evolving graphs significantly.