M-tree: An Efficient Access Method for Similarity Search in Metric Spaces

A new access meth d, called M-tree, is proposed to organize and search large data sets from a generic "metric space", i.e. whE4 object proximity is only defined by a distance function satisfyingth positivity, symmetry, and triangle inequality postulates. We detail algorith[ for insertion of objects and split management, whF h keep th M-tree always balanced - severalheralvFV split alternatives are considered and experimentally evaluated. Algorithd for similarity (range and k-nearest neigh bors) queries are also described. Results from extensive experimentationwith a prototype system are reported, considering as th performance criteria th number of page I/O's and th number of distance computations. Th results demonstratethm th Mtree indeed extendsth domain of applicability beyond th traditional vector spaces, performs reasonably well inhE[94Kv#E44V[vh data spaces, and scales well in case of growing files. 1