Adaptive Implementation to Update Page Rank on Dynamic Networks

Page Rank is a very ubiquitous network metric for large networks, such as the world-wide-web, ranking the vertices based on this metric gives an idea about its importance and quality. There are a lot of attempts made to parallelize the page rank algorithm for static networks, however, there are only very few attempts made to compute page rank on dynamic networks. As the networks change with time, computing page rank or updating is an expensive operation, the previous attempts have only approximated the metric to avoid recomputation. In this paper, we introduce a framework where we try to update the page rank of the vertices which embraces change as the network changes. The proposed framework is implemented on a shared memory system and experiments on real-world and synthetic networks show good scalability. The framework proposed gets an input set of network, initial page rank values for all the vertices, and a set of batches that has the changeset. As the batches are processed in parallel, affected vertices are identified and marked for an update, once the batch is processed the vertices affected or identified their page rank values are computed. The main contribution of this paper is the proposed framework avoids recomputation of all vertices, and only recomputes few vertices, and avoids approximation to provide accurate values.