LIFT:一种用于高维数据的索引结构

本文提出一种新的高维空间中点数据的索引方法,其基本原理是用格矢量量化(Lattice vector quantization)均匀划分数据空间、用倒排文件(Inverted File)存储格点、用Trie树实现倒排文件的组织和存储、用Trie并行搜索算法实现倒排文件的快速访问.和传统索引方法相比,新方法具有许多优点,例如它能以较低的复杂度建立索引结构、支持非常高维的数据索引、充分利用高维空间中点分布的稀疏性等.实验结果表明,在较高维数时,LIFT性能优于传统索引方法.

LIFT:An Index Structure for High Dimensional Data

A new method for indexing large amounts of points in high-dimensional space is proposed.The basic principle is as follows:uniformly partition the data space by Lattice vector quantization,store the lattice points by Inverted File,organize the inverted file by Trie tree,and fast access the inverted file by Trie parallel search algorithm.We called this index structure LIFT.Compared with the traditional index methods,the LIFT can build the index structure with low complexity,support very high dimensionality,and take advantage of sparsity of data points in high-dimensional space,etc.The experiments show that for high-dimensional data,the LIFT outperforms the well-known R-tree.