With LSA size DOES matter

Abstract

Latent Semantic Analysis (LSA) is a technique from the field of Natural Language Processing that enables comparison of semantic similarities between documents using vector operations. This technique has been used in areas from Information Retrieval (IR) to the automated assessment of essays. One property used in document comparison is size. The general philosophy is that more text is better although few concrete examples or guidelines exist that demonstrate this. This paper shows, via a novel concrete example taken from real world data, that larger documents do imply more accurate semantic similarity comparisons.