Efficient Exact Learning Algorithms for Road Networks and Other Graphs with Bounded Clustering Degrees

Abstract

The completeness of road network data is significant in the quality of various routing services and applications. We introduce an efficient randomized algorithm for exact learning of road networks using simple distance queries, which can find missing roads and improve the quality of routing services. The efficiency of our algorithm depends on a cluster degree parameter, d max , which is an upper bound on the degrees of vertex clusters defined during our algorithm. Unfortunately, we leave open the problem of theoretically bounding d max , although we conjecture that d max is small for road networks and other similar types of graphs. We support this conjecture by experimentally evaluating our algorithm on road network data for the U.S. and 5 European countries of various sizes. This analysis provides experimental evidence that our algorithm issues a quasilinear number of queries in expectation for road networks and similar graphs. of → Graph analysis; Theory of → Random network models;