Analysis of Semantic Space of Texts and its Application to Psychometrics

Cheongtag Kim

Abstract

   Most of the measurements used in experimental psychology are response times and accuracies produced by well-designed experiments. Recently, a big amount of text data acquired in natural setting are accessible and usable. We believe that analysis of this type of data will expand the understanding of human cognition.

   This presentation inquires into the ways of using text analysis in the domain of cognition and individual differences. For the cognitive domain, we discuss how semantic structure of text can be discovered by analyzing document-term matrix. Various types of latent semantic analysis are practiced to affirm that the text data hold the information about human semantic processes and it can be uncovered by the text analyses. We also demonstrated how these semantic analyses can be used to text categorization and detect the meaning of ambiguous words. For the individual difference (psychometrics) domain, we applied these techniques to find relevant items when constructing test questionnaires. By analyzing the text produced by subjects who was instructed to write short descriptions about several constructs, a set of words can be extracted that can be used as items for each construct. Finally, some possible applications of text analyses are discussed.