R-trees are widely used in spatial and multi-dimensional databases. However, according to our investigation, the overlap among the leaf nodes of R-trees is serious and the objects are not well-clustered in the leaf nodes, which greatly affect the effect of the pruning strategies when nearest neighbor searching is performed and also affect the other search performance of R-trees. The forced reinsertion introduced in R*-tree can improve this problem to some extent, but can not completely solve this problem. In this study, we try to combine SUM (Self Organizing Map) technology and R*-tree technology to lessen the overlap among the leaf nodes of R*-tree and to improve the clustering degree of the objects in the leaf nodes. The experimental result shows that the SUM-based R*-tree proposed in this paper has a much better search performance than R*-tree.
|Translated title of the contribution||A New SOM-based R*-tree: Building and Retrieving|
|Number of pages||6|
|Journal||Research Reports on Information Science and Electrical Engineering of Kyushu University|
|Publication status||Published - Sep 2001|