1 Intro to Latent Semantic Analysis

The purpose of this document is to explain why LSA works – specifically, why (and when) is it (mathematically) justified to use the similarity between term vectors or document vectors. All of the material here appears in the highly cited paper “Indexing by Latent Semantic Analysis ” [1] (as well as other publications introducing the LSA and LSI methods). However, it is unfortunately not discussed much in NLP publications that use it