摘要
在文本聚类中,相似性度量是影响聚类效果的重要因素。常用的相似性度量测度,如欧氏距离、相关系数等,只能描述文本间的低阶相关性,而文本间的关系非常复杂,基于低阶相关测度的聚类效果不太理想。一些基于复杂测度的文本聚类方法已被提出,但随着数据规模的扩展,文本聚类的计算量不断增加,传统的聚类方法已不适用于大规模文本聚类。针对上述问题,提出一种基于MapReduce的分布式聚类方法,该方法对传统K-means算法进行了改进,采用了基于信息损失量的相似性度量。为进一步提高聚类的效率,将该方法与基于MapReduce的主成分分析方法相结合,以降低文本特征向量的维数。实例分析表明,提出的大规模文本聚类方法的聚类性能比已有的聚类方法更好。
Text clustering is the key technology of text organization,information extraction and topic retrieval.Appropriate similarity measure selection is an important task of clustering,which has great affection on the clustering results.Classical similarity measures,such as distance function and the correlation coefficient,can only describe the linear relationship between documents.However,clustering results based on classical clustering methods are usually unsatisfactory due to the complicated relationship among text documents.Some complicated clustering methods have been studied.But,with the growing scale of text data,the computational cost increases markedly with the increase of dataset size.Classical clustering methods are out of work in dealing with large scale dataset clustering problems.In this paper,a distributed clustering method based on MapReduce was proposed to deal with large scale text clustering.Furthermore,we proposed an improved version of k-means algorithm,which utilizes information loss as the similarity function.For improving clustering speed,parallel PCA method based on MapReduce was used to reduce the document vector dimension.The experimental results demonstrate that the proposed method is more efficient for text clustering than classic clustering methods.
出处
《计算机科学》
CSCD
北大核心
2016年第1期246-250,269,共6页
Computer Science
基金
国家自然科学基金项目(61472230)
山东省科技发展计划(2013GZC20102)资助
