Optimizing Similarity Search in the M-Tree

A topic of growing interest in a wide range of domains is the similarity of data entries. Data sets of genome sequences, text corpora, complex production information, and multimedia content are typically large and unstructured, and it is expensive to compute similarities in them. The only common denominator a data structure for eicient similarity search can rely on are the metric axioms. One such data structure for eicient similarity search in metric spaces is the M-Tree, along with a number of compatible extensions (e.g. Slim-Tree, Bulk Loaded M-Tree, multiway insertion M-Tree, M2-Tree, etc.). The M-Tree family uses common algorithms for the k-nearest-neighbor and range search. In this paper we present new algorithms for these tasks to considerably improve retrieval performance of all M-Tree-compatible data structures.