基于知网语义相似度计算的特征降维方法研究

针对文本分类处理中的高维度问题,结合知网语义词典,提出了一种新的特征降维处理方法.通过计算特征词汇之间的语义相似度,将原有特征集分成若干特征词集;同一词集内的特征词语义彼此间相似;而不同词集的特征词彼此间相似度比较小.将同一特征词集内的词汇权重相加,从而突出同义词以及近义词对文本分类的贡献,并可以大大降低文本比较的特征维数.实验结果表明,利用该方法在文本分类中得到了较好的分类准确率和分类性能.

The Research on Reduced Feature Dimension Based on Hownet Similarity Computing

Aimed to solve high dimension in text classification, a new reduced feature dimension measure is provided combining with Hownet semantic dictionary. The semantic similarity among feature glossary is firstly analysised, then the semantic similar features into subclusters is put, which make the primary feature set classified into several subclusters. Finally it condenses the subclusters to reduce feature dimension in text comparison. The experimental results show that the new method can get better performance in text classification.