关键词: 潜在语义索引, 文本挖掘, 向量空间模型, 奇异值分解

Abstract: This paper introduced the application of latent semantic indexing in the field of information retrieval, and presented three ways to calculate the lexical item weighting, and then analyzed the role of Singular Value Decomposition (SVD) in capturing the important information of matrix, and showed how the reduced-rank approximation of item-document matrix simulated the psychological process of human when understanding the meanings of sentences. Through the comparison of the searching algorithm of Vector Space Model (VSM) and LSI, and the case of text mining of a term-document matrix, it indicated how LSI worked in analyzing the connection between documents.

Key words: Latent semantic indexing