基于KNN+层次SVM的文本自动分类技术  

INTEGRATING KNN AND HIERARCHICAL SVM FOR AUTOMATIC TEXT CLASSIFICATION

作　　者：王金华[1] 喻辉[2] 产文 周向东[3] 施伯乐[3] 

机构地区：[1]中国电子科技集团公司第三十二研究所,上海200233 [2]成都军区通信网络技术管理中心,四川成都610000 [3]复旦大学计算机学院,上海200433

出　　处：《计算机应用与软件》2016年第2期38-41,共4页Computer Applications and Software

摘　　要：针对大规模文本的自动层次分类问题,K近邻(KNN)算法分类效率较高,但是对于处于类别边界的样本分类准确度不是很高。而支持向量机(SVM)分类算法准确度比较高,但以前的多类SVM算法很多基于多个独立二值分类器组成,训练过程比较缓慢并且不适合层次类别结构等。提出一种融合KNN与层次SVM的自动分类方法。首先对KNN算法进行改进以迅速得到K个最近邻的类别标签,以此对文档的候选类别进行有效筛选。然后使用一个统一学习的多类稀疏层次SVM分类器对其进行自上而下的类别划分,从而实现对文档的高效准确的分类过程。实验结果表明,该方法在单层和多层的分类数据集上的分类准确度比单独使用其中任何一种要好,同时分类时间上也比较接近其中最快的单个分类器。For automatic hierarchical classification of large-scale text, k-nearest neighbours （KNN） algorithm has higher classification efficiency but is not effective for classifying the samples on the borders of categories. The support vector machine （SVM） classification algorithms have higher accuracy, however a number of previous multi-class SVM algorithms are composed of a number of independent binary classifiers, thus they become slower in training process and are not suitable for hierarchical category structures. This paper presents a new method which integrates both KNN and hierarchical SVM algorithm for automatic text classification. First we modify the KNN algorithm to quickly obtain K class labels of the nearest neighbours, and effectively sift out candidate categories of the documents with them. Then we use a multi-class sparse hierarchical SVM classifier with uniform learning to make top-down categories partition on the sample, so that implement the efficient and accurate classification process on the documents. Experimental results demonstrate that the classification accuracy of this method on classification dataset with single-layer and multi-layer is better than just using either of the methods, meanwhile it is also close to the fastest single classifier in classification time.

关 键 词：自动文本分类 KNN 层次SVM 

分 类 号：TP302.1[自动化与计算机技术—计算机系统结构]