基于SOM聚类的软构件分类方法

Software Component Classification Based on Improved SOM Clustering

软构件刻面分类法是一种被各大软构件库系统广泛采用的分类方法,但是传统的刻面分类法需要人工建立和维护庞大的术语空间,增大了软构件建库和入库的工作量。利用基于SOM神经网络的聚类技术可实现无需建立术语空间的软构件自动分类,同时针对软构件的特点和SOM聚类的需要预先确定拓扑结构和聚类结果与输入样本的次序有关等缺点,对SOM聚类的训练过程进行改进以满足软构件聚类的要求。

Faceted classification is a popular software component classification adopted by many software component repository systems. However, it needs artificially establish term space of each facet, which increases the workload of establishing the software component repository and the workload of inserting components into the repository. Based on SOM clustering, an automatic component classification is provided to free faceted classification from term space. SOM clustering has two disadvantages such that their topology construction need be defined in advance and the cluster result is disturbed by order of learning samples. So the training process of SOM clustering is improved to increase the accuracy of software component classification.