改进的K最近特征线算法在文本分类中的应用  

Improved K Nearest Feature Line Algorithm in Text Categorization

作　　者：谭冠群[1] 丁华福[1] 

机构地区：[1]哈尔滨理工大学计算机科学与技术学院,黑龙江哈尔滨150080

出　　处：《哈尔滨理工大学学报》2008年第6期19-22,共4页Journal of Harbin University of Science and Technology

摘　　要：KNFL算法是一种近年来在人脸识别领域得到广泛应用的算法,这种算法认为类中两点的连线也可以近似代表类的特征,把它应用于文本分类领域可以得到较好的分类效果,但是由于时间复杂度比较高,影响了其实用价值.本文提出了一种应用于文本分类的改进的KNFL算法,计算出类的中心点后再进行两次过滤,分别将离类中心点较远的特征点和特征线过滤掉,减少了训练集样本数目,在对分类精确度影响不大的情况下,改善了KNFL算法的分类效率,最后用实验验证了该算法的有效性.KNFL algorithm has been a popular classification algorithm in face identification in recent years. This algorithm considers that a line between two points in the same type of class can nearly represent the feature of the whole class as a single point, and it can also get good results in text categorization field, but its computational complexity is in high degree which impact its valve for application. A novel and efficient improved KNFL algorithm for text categorization is proposed in this text, computering the central point of class and then making two filters, deleting the points and the feature lines whose distance of the central point is far, as a result we reduce the number of the point in training set. This algorithm saves the classification time and has the same classification performance as the KNFL algorithm. At last experimental results based on text categorization will verify our proposed algorithm.

关 键 词：文本分类 K最近特征线 KNN算法 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]