基于图划分的谱聚类算法在文本挖掘中应用

Application of Spectral Cluster Algorithm Based on Graph Partition in Text Mining

传统文本挖掘算法都是建立在凸球形的样本空间上,当样本空间不为凸时,算法就陷入"局部"最优。为了满足"全局"最优,引进了无向图结构表示文档之间的相似关系,由无向图建立文档之间的相邻接矩阵,谱聚类算法是通过对邻接矩阵进行分析,导出聚类对象的新特征,利用新的特征对原数据进行聚类。通过实验对该算法和其他的文本挖掘的算法进行分析比较,实验结果表明该算法聚类效果比传统数据挖掘方法好。最后指出谱聚类的不足和进一步的研究方向。

Traditional text mining algorithms are based on the sample of spherical convex space, when sample space is not convex, the algorithm performance on a ＂local＂ optimization, introduced no direction graph to map similar relationship between documents, then set up adjacent matrix between documents, spectral clustering algorithm analysis adjacent matrix to get the new cluster features, then use the new features to cluser,then it is compared with other algorithm of the text mining by experimental methods. The results showed that the spectral duster algorithm produced good effect. At last, descripts shortage and further research directions of spectral cluster algorithm.