基于模糊相似度的RPCL文本聚类算法

文本聚类过程中,存在着文本数据空间维数巨大,聚类的数目不能直接确定等问题。为此,有专家学者提出了次胜者受罚的竞争学习(Rival Penalized Competitive Learning)算法,简称RPCL算法。该算法在一定程度上,解决了聚类的数目的确定问题。但是,该算法只适合做低维数据的聚类,对于高维数据聚类效果极差。该文提出了一种改进的RPCL算法,该方法不再采用欧氏距离去计算相似度,而是采用模糊相似度的方法,通过实验表明,改进的RPCL算法在聚类效果上好于经典的RPCL算法。

RPCL Text Clustering Based on Fuzzy Similarity Degree

There exist some problems in text clustering,such as huge dimensionality in text feature matrix,unknown cluster number.Some expert put forward the algorithm of Rival Penalized Competitive Learning,RPCL for short.The algorithm solved the problem of determining the clustering number in some degree.However,it is suited to cluster of text in lower dimension,not to higher dimension.This paper introduces a improved RPCL algorithm.Fuzzy similarity instead of Euclid distance is used to compute similarity.The experi...