Test Case Selection for Evaluating Measures of Semantic Distance
Wayne D. Gray, Vladislav D. Veksler · eScholarship (California Digital Library) · 2006
With a growing number of measures of semantic distance (MSD), such as LSA, GLSA, PMI-IR, WordNet, Normalized Google Distance (NGD), etc., we find the need for a standard direct comparison of these algorithms.A oneclick browsing task affords direct comparisons of two or more MSDs.In the simplest version of the task, subjects search for a target term, T, and have to choose between one of two links, A and B. A choice of link A over link B suggests that in the subject's mind, T is more similar to A than it is to B.Here we formalize a procedure to select test case triads T-A-B, such that a human choice of A over B would directly implicate that one of the MSDs being investigated accords better with human judgment than another.