基于角相似性的k最近邻搜索研究

Angular Similarity Based K-nearest Neighbor Search

在高维空间中k最近邻搜索（KNNS）应用非常广泛，但是目前很多KNNS算法都根据欧氏距离对数据进行索引和搜索，不适合采用角相似性的应用。本文提出一种基于角相似性的k最近邻搜索算法（AS—KNNS）。该算法先提出基于角相似性的数据索引结构（AS-Index），参照一条中心线和一条参照线，将数据以系列壳．超圆锥体方式进行组织并分别线性存储；然后确定查询对象的空间位置，有效确定一个以从原点到查询对象的直线为中心线的超圆锥体并在其中进行搜索。实验结果表明，AS-KNNS算法较其他k最近邻搜索算法有更好的性能。

The k-nearest search algorithm（KNNS） is widely used in the high dimension space. However, the current KNNS uses Euclidean distance to index dataset and retrieve the target object, which is not suitable for those applications based on angular similarity. In this paper, the angular similarity based on KNNS （AS-KNNS） is proposed. AS-KNNS firstly proposes that the indexing structure should be based on angular similarity, refer to a center line and a referenced line to organize dataset with the method of the shell-hypercone, and store them linearly. Then it determines the space place for the target object, making a hypercone which takes the line connecting the origin point and the target object as center, and searches the hypercone for the target. The experiment shows that the performance of AS-KNNS is superior to those other KNNS.