运用语义集索引法实现英文文本分类

Realizing English Text Classification with Semantic Set Index Method

为克服当前文本分类法中基于词形匹配带来的局限性,基于WordNet语义词典和隐含语义索引(LSI)模型,提出了基于语义集索引的英文文本分类方法.该方法在分类初期首先利用WordNet构建语义词典库,利用单词的语义集代替单词作为文本特征向量的特征项;然后利用LSI模型进一步深入挖掘语义集概念间的深层联系,将语言知识和概念索引有效地融合到文本向量空间的表示中.针对Na ve Bayes及简单向量距离文本分类法的实验结果显示,2种文本分类法的分类准确率均随着语义分析的深入逐步提高,充分表明了语义挖掘对文本分类的重要性和必要性.

To overcome the limitations of actual text classification methods based on bag-of-words representation, An English text classification method based on semantic set index is presented from the WordNet thesaurus and LSI （latent semantic indexing） model. At the initial stages of text classification, the method first constructs semantic thesaurus database by WordNet and replaces bag-of-words with bag-of-semantic sets as an element of the text feature vector. Then LSI model will be used to further mine the deep-seated relations among concepts represented by semantic sets. It effectively incorporates linguistic knowledge and conceptual index into text vector space representation. The experimental results aiming at Naive Bayes and simple vector distance text classification methods show that the accuracy rates of the two classification methods are gradually improved along with more and more in-depth semantic analysis, fully indicating that semantic mining is very important and necessary to text classification.