基于论域划分的无监督文本特征选择方法

由于缺乏类信息,使得无监督文本特征选择问题一直未较好地加以解决。为此,对该问题进行了研究并提出了一个基于论域划分的无监督文本特征选择。该方法主要是把论域划分的思想引入到无监督文本特征选择之中,其首先使用一种新型无监督文档进行文本特征初选以过滤低频的噪声词,然后再使用所给的基于论域划分的属性约简进行文本特征优选。实验结果表明这个方法能够克服文本聚类时缺乏类的先验知识的不足,可以较好地解决无监督文本特征选择问题。

Unsupervised Text Feature Selection Method Based on Domain Division

Due to the lack of class labels, unsupervised text feature selection problem hasn't been resolved effectively. Therefore, this problem was studied in this thesis and an unsupervised text feature selection method based on domain division was proposed. This method mainly makes use of supervised text feature selection doing unsupervised text feature selection. The proposed firstly uses a new unsupervised Document Frequency filter out those low-frequency noise words, and then employs a presented attribute reduction based on domain division for text feature optimization. The experimental results show that this method can overcome the clustering flaw which lacks of transcendent knowledge and solve unsupervised text feature selection problem well.