Abstract: Feature weight algorithm has great impact on the classification results. Traditional algorithms don’t consider distribution information among and inside classes. This paper introduces a new improvement ideas of skew information among classes, distribution information inside a class and weight adjustment factor, then puts forward a new term weight algorithm based on WA-DI-SI after in depth analysis of improvements method, and uses SVM and na?ve Bayesian classifier to check its validation. The method is better than others and proves that the improved algorithm is feasible.

Key words: text classification, feature weight, skewness among classes, dispersion inside a class, weight adjustment factor

摘要： 特征权重算法对文本分类系统的精确度有很大影响，传统的TFIDF算法未能考虑特征项在类间和类内的分布情况。为此，在对传统算法和相关改进算法进行分析的基础上，引入类间偏斜度、类内离散度和权重调整因子的改进思路，提出一种基于WA-DI-SI的特征权重改进算法，分别采用支持向量机和朴素贝叶斯2种分类算法进行测试。测试结果表明，与其他改进算法相比，该算法能够获得更好的分类效果。

关键词: 文本分类, 特征权重, 类间偏斜度, 类内离散度, 权重调整因子

CLC Number: 

• TP311.52