基于概念词的特征提取方法  

Feature Extraction Method Based on Concept-word

作　　者：何杰[1] 朱征宇[1] 董曙佳[1] 于春雷[1] 

机构地区：[1]重庆大学计算机学院,重庆400044

出　　处：《世界科技研究与发展》2012年第1期119-122,147,共5页World Sci-Tech R&D

基　　金：国家科技支撑计划课题-重庆便民e站服务平台(2007BAH08B04)资助项目

摘　　要：为解决因未考虑语义关联造成的VSM描述不准确的问题,基于知网本体库计算词语间的语义相似度,采用识别完全子图的方式生成概念词列表,再用概念词替换存在密切语义关联的词语。实验表明,该方法在改进文档特征提取效果的同时也明显降低了向量空间的维度。与不经概念词处理的特征提取方法相比,该方法在分类识别率上有一定提升。In order to solve the problem in the inaccurate description of the Vector Space Model, a feature extraction method is proposed basing on Concept-word, considering the semantic association between words. Firstly, the semantic similarity between words is calculated basing on the HowNet, From the similarity list, the complete subgraph recognition is taken to generate a list of Concept-words. Then words of closely related are replaced with the Concept-words. The effect of the document extraction is improved. The dimensions of document vector are re- duced. The results show that the accuracy of classification is improved, compared with the method without Concept-word dealt.

关 键 词：概念词 知网 向量空间模型 特征提取 

分 类 号：TQ316.241[化学工程—高聚物工业]