Decentralized Search for Shortest Path Approximation in Large-Scale Complex Networks

Finding approximated shortest paths for extremely large-scale complex networks is a challenging problem, where existing works require large overhead to achieve high accuracy and diversity for estimated paths, especially for large graphs with millions of vertices. In this paper, we propose an online search approach based on preprocessed indexes, to approximate point-to-point shortest paths. The approach is able to find more accurate and diverse paths with limited index overhead and requires low search overhead. Furthermore, a new path degree based index construction algorithm is introduced that can greatly increase the approximation accuracy and involve no additional index overhead. To handle extreme size graphs, we build a query processing system with our algorithm on distributed graph processing platforms. The system also supports parallel processing of online searches to achieve high throughput for a large number of queries. We evaluate our algorithm on various real-world graphs from different disciplines with up to billions of edges, and we demonstrate that our system can process hundreds of thousand queries per second on these graphs with reducedoverhead.