基于语义与最大匹配度的短文本分类研究  

Short text classification based on semantics and maximum matching degree

作　　者：孙建旺[1] 吕学强[1] 张雷瀚[1] 

机构地区：[1]北京信息科技大学网络文化与数字传播北京市重点实验室,北京100101

出　　处：《计算机工程与设计》2013年第10期3613-3618,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61271304);北京市教委科技发展计划重点项目暨北京市自然科学基金B类重点项目(KZ201311232037)

摘　　要：为了解决基于VSM方法在进行短文本分类时存在的严重数据稀疏问题,提出了基于语义与最大匹配度的短文本分类方法。以《知网》为知识源,设计了基于义原距离、义原深度与区域密度的义原相似度计算方法,实现基于词类的词语相似度计算;提出了基于语义与最大匹配度的方法计算短文本相似度,应用KNN算法进行短文本分类。实验结果表明,该方法与基于语义、基于AD_NB等方法相比,正确率、召回率和F值均得到了明显的提高。To deal with the serious data sparseness problem exists in the traditional VSM method of carrying out short text classification,a short text classification method based on the semantics and maximum matching degree is put forward.The primary similarity calculation method is designed based on the distance,the depth and area density.Word similarity calculation is carried out according to its part of speech and HowNet is utilized as a source of knowledge.The short text similarity calculation based on the method of combining the semantics and maximum degree is proposed.KNN algorithm is applied to the short text classification.Experimental results show that the precision,recall and F-measure are significantly improved in contrast with those of the method based on the semantics,AD_ NB and so on.

关 键 词：短文本分类 义原相似度 词语相似度 语义 最大匹配度 KNN算法 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]