Latent Semantic Analysis: five methodological recommendations

Abstract

The recent influx in generation, storage and availability of textual data presents researchers with the challenge of developing suitable methods for their analysis. Latent Semantic Analysis (LSA), a member of a family of methodological approaches that offers an opportunity to address this gap by describing the semantic content in textual data as a set of vectors, was pioneered by researchers in psychology, information retrieval, and bibliometrics. LSA involves a matrix operation called singular value decomposition, an extension of principal component analysis.LSA generates latent semantic dimensions that are either interpreted, if the researcher's primary interest lies with the understanding of the thematic structure in the textual data, or used for purposes of clustering, categorisation and predictive modelling, if the interest lies with the conversion of raw text into numerical data, as a precursor to subsequent analysis. This paper reviews five methodological issues that need to be addressed by the researcher who will embark on LSA. We examine the dilemmas, present the choices, and discuss the considerations under which good methodological decisions are made. We illustrate these issues with the help of four small studies, involving the analysis of abstracts for papers published in the European Journal of Information Systems.

Quantitative approaches to content analysis: identifying conceptual drift across publication outlets

“…Therefore, although our proof of concept highlights the A U T H O R C O P Y ability to identify longitudinal conceptual drift, we point out that both LSA and Leximancer can be applied to a wide variety of other analyses of any type of textual data. For instance, we point the interested reader to the five methodological recommendations pertaining to the use of LSA described by Evangelopoulos et al (2011).…”

The Tale of e-Government: A Review of the Stories that Have Been Told So Far and What is Yet to Come

“…This situation resulted from the ever-growing appearance and increased use of web-and cloud-based systems, mobile devices and social media as well as the online storage possibilities of information. New ways of analyzing this massive information like text mining have therefore received growing attention in the last years and have already been successfully applied in research, including the IS discipline [12,17,20,47]. In this respect, this approach represents a powerful tool, because it allows to go beyond the natural boundaries of manual data analyses (quantity-wise), and quasi de facto excludes humanly induced biases (content-wise) [46].…”

Feedback recurrent neural network-based embedded vector and its application in topic model

“…With these term frequency vectors, researchers could compute the similarities between documents and thus discover underlying topics in such a document collection [2]. Topic models can be classified into statistical semantic models [3][4][5][6][7][8][9][10] and embedded vector models [11][12][13]. While capturing the semantics of documents, statistical semantic model computes the similarities between documents with co-occurrence matrix of terms, and embedded vector model uses neighbor(s) to represent the meaning of a target term; however, both of them cannot describe the term orders in a document.…”