A technique for evaluating automatic term clustering
Martin C. Dillon, P. Caplan · Journal of the American Society for Information Science · 1980
Abstract This study reexamines the feasibility of automatic thesaurus generation by establishing a framework for analyzing the degree of correspondence between automatically generated thesauri and a thesaurus of the same terms constructed by a team of subject specialists. In order to test the evaluative procedure, a series of thesauri are generated using the same clustering algorithm and three different measures of association. By comparing the relative effectiveness of the generated thesauri to the human standard, it is determined that substantial differences exist among them.