摘要
传统tf.idf算法中的idf函数只能从宏观上评价特征区分不同文档的能力,无法反映特征在训练集各文档以及各类别中分布比例上的差异对特征权重计算结果的影响,降低文本表示的准确性。针对以上问题,提出一种改进的特征权重计算方法tf.igt.igC。该方法从考察特征分布入手,通过引入信息论中信息增益的概念,实现对上述特征分布具体维度的综合考虑,克服传统公式存在的不足。实验结果表明,与tf.idf.ig和tf.idf.igc 2种特征权重计算方法相比,tf.igt.igC在计算特征权重时更加有效。
The idf function of traditional (f..idf algorithm can only evaluate the ability of features to discriminate different documents in a macroscopically way, which can not reflect the differences of distribution proportion for features in each document and each class of the whole training set, it reduces the accuracy of text representation. To solve the above problem, this paper proposes an improved feature weighting method called tfig,.igc. This method begins from analyzing the characteristics of feature distribution, through introducing the concept of information gain in the information theory, realizes the comprehensive consideration of the two specific dimensions of feature distributions, and overcomes the shortcomings of the traditional formula. Experimental results on the two open source corpus show that compared to other two feature weighting methods, tf.ig.igc is more effective in terms of calculating the feature weighting.
出处
《计算机工程》
CAS
CSCD
北大核心
2011年第1期16-18,21,共4页
Computer Engineering
基金
中国博士后科学基金资助项目(20090461425)
江苏省博士后科研计划基金资助项目(0901014B)
关键词
特征分布
特征加权
文本分类
feature distribution
feature weighting
text classification