Latent Semantic Analysis Captures Casual, Goal-oriented, and Taxonomic Structures
Arthur C. Graesser, Ashish Karnavat, Victoria Pomeroy, Katja Wiemer-Hastings · eScholarship (California Digital Library) · 2000
Latent Semantic Analysis (LSA) has been used to represent the domain of computer literacy in AutoTutor, a fully automated computer tutor.The analyses in the present study support the claim that the 200-dimensional LSA space captures aspects of the structured mental models that underlie computer literacy.Knowledge structures were constructed that contained causal networks, goal/plan/action hierarchies, and taxonomic hierarchies.The proximity of a pair of nodes (i.e., concept, state, event, action, goal) in these structures predicted the cosine similarity scores that are routinely computed in LSA analyses.