基于KNN与自动检索的迭代近邻法在自动分类中的应用

本文研究了一种基于KNN与自动检索的自动分类算法———迭代近邻法 (IterativeKNN ,I KNN) ,用以解决KNN算法在小样本库的环境下分类效果不佳的问题。在无法得到足够的定类样本时 ,通过检索的方法将待分样本的局部主题特征放大 ,进而得到足够定类的相似样本。实验证明 ,迭代近邻法既增加了获取相似样本的几率 ,同时也有效地控制了样本相似度条件限制放宽后可能引入的分类噪声 ,在实际应用中能较好地提升自动分类系统的查全率和查准率。

Application of Iterative-KNN Based on KNN and Automatic Retrieval in Automatic Categorization

In this paper, an approach called I KNN, based on KNN and automatic retrieval, is proposed in order to improve the performance of KNN algorithm with a small scale document database. If we cannot get enough similar documents in such condition, we magnify part of the document subjects to find more similar documents by Internet retrieval. It is proved by experiment that I KNN can improve the recall and precision of automatic categorization system.