基于改进LSA的文档聚类算法  

Document Clustering Algorithm Based on Modified Latent Semantic Analysis

作　　者：俞辉[1] 

机构地区：[1]中国石油大学计算机与通信工程学院,山东东营257061

出　　处：《小型微型计算机系统》2009年第5期963-966,共4页Journal of Chinese Computer Systems

摘　　要：提出一种基于改进潜在语义分析MLSA(Modified Latent Semantic Analysis)的文档聚类算法.采用新的特征提取方法构建词-文档矩阵,利用潜在语义分析对词-文档矩阵进行奇异值分解以达到垃圾信息过滤的目的,同时使得向量空间模型中文本的高维表示变成在潜在语义空间中的低维表示,缩小了问题的规模.然后将共现数据对转换成概率统计模型来计算,提高了聚类质量.实验表明,本文提出的方法是有效的.This paper proposed a new algorithm of document clustering based on modified latent semantic analysis. New method of feature extraction was used to construct word-document matrix. Latent semantic analysis which stems from linear algebra performed a Singular Value Decomposition of word - document matrix, so that not important information was filtered, and the high dimension represent of document in Vector Space Model was changed to low dimension represent in latent semantic space. Co-occurrence data was changed to probabilistic model by modified latent semantic analysis, the performance of clustering was improved. Experimental result shows that the proposed cluster algorithm is effective.

关 键 词：文档聚类 潜在语义分析 奇异值分解 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]