Lossy Graph Data Reduction

Graphs are widely used nowadays to store complex data of large applications such as social networks, recommendation engines, computer networks, bio-informatics, just to name a few. Graph data reduction plays a very important role in order to store and process such data efficiently. Many studies about graph data reduction have been done, but most of them are focused on lossless reduction in the sense that query results are preserved after reduction. In this paper, we elaborate on the concept of "lossy" graph reduction for applications that may tolerate approximate results with small but bounded errors in exchange for further data reduction. We study one well known graph problem that is the shortest path problem and design the lossy graph reduction algorithms. The error bounds of the algorithms we propose are proved theoretically. In addition, we implement some of the algorithms based on real world data sets to experimentally investigate the impacts of the error tolerance on the reduction ratio.