基于幂迭代-随机奇异值分解的潜在语义分析  

Latent semantic analysis based on power iteration-randomized singular value decomposition

作　　者：黄斯怡 陈兴荣[1] HUANG Siyi;CHEN Xingrong(China University of Geosciences,School of Mathematics and Physics,Wuhan 430074,China)

机构地区：[1]中国地质大学(武汉)数学与物理学院,湖北武汉430074

出　　处：《厦门大学学报（自然科学版）》2023年第4期679-686,共8页Journal of Xiamen University：Natural Science

摘　　要：潜在语义分析通过挖掘文本与单词之间基于话题的关系,进行文本情感分类.潜在语义分析通过截断奇异值分解实现,传统截断奇异值需要求解大型矩阵的特征值和特征向量,存在时间复杂度高的问题.利用幂迭代-随机奇异值分解代替传统奇异值分解,提出一种矩阵分解效率更高的潜在语义分析算法.为了检验改进算法的性能,将其与四种分类算法组合,用于四个评论数据集的文本情感分类实验.结果显示,在保持原有较高分类准确率的基础上,改进的潜在语义分析算法的计算效率更高.Latent semantic analysis mines the topic-based relationship between the text and words to achieve the purpose of text sentiment classification.They are implemented by truncated singular value decomposition.Traditional truncated singular value decomposition needs to solve the eigenvalues and eigenvectors of large matrices,thus enduring the problem of high time complexity.By means of power iteration-randomized singular value decomposition that replaces traditional singular value decomposition,a latent semantic analysis algorithm with higher matrix decomposition efficiency is proposed.To test the performance of the improved algorithm,we combine this algorithm with four classification algorithms for text sentiment classification experiments in four comment datasets.Experimental results show that,while maintaining high classification accuracies,the proposed algorithm performs efficiently.

关 键 词：文本情感分类 潜在语义分析 奇异值分解 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]