Efficient AKNN spatial network queries using the M-Tree

Abstract

Aggregate K Nearest Neighbor (AKNN) queries are problematic when performed within spatial networks. While simpler network queries may be solved by a single network traversal search, the AKNN requires a large number costly network distance computations to completely compute results. The M-Tree index, when used with Road Network Embedding, provides an efficient alternative which can return estimates of the AKNN results. The M-Tree index can then be used as a filter for AKNN results by quickly computing a superset of the query results. The final AKNN query results can be computed by sorting the results from the M-Tree. In comparison to Incremental Euclidean Restriction (IER), the M-Tree reduces the overall query processing time and the total number of necessary network distance computations required to complete a query. In addition, the M-Tree filtering method is tunable to allow increasing performance at the expense of accuracy, making it suitable for a wide variety of applications.