Twisting the Metric Space to Achieve Better Metric Trees

In order to index text fields in XML databases for similarity queries, M-Trees may be applied. However, some datasets experiments using M-Trees for this purpose resulted in low query performance. The problem is that the arrangement of the objects in these datasets over the metric space seems to be inappropriate for grouping. This can jeopardize query performance because the M-Tree structure is based upon grouping of similar objects. We propose applying a twisting function over the metric space to generate a new space, the "twisted space". In this new space, objects are better arranged enabling a more appropriate grouping. However, this twisted space cannot be used for querying, because it does not have the metric properties required by M-Trees. Therefore, we need to use both, twisted and metric space. The paper presents such modified M-Trees as well as experiments that show performance improvements.

• Type: Conference
• Year: 2004
• Where: SBBD
• Authors: César Feijó Nadvorny, Carlos A. Heuser
• Keywords: Metric Space, Query Performance, SBBD 2004, SBBD 2007, Twisted Space