The Superiority of Relative Criteria in Partial Matching and Generalization

Abstract

Some familiar justifications for concluding that a data element D is an unacceptable match to a pattern element P are examined for their suitability for partial-matching applications. All of these absolute criteria are found to be unacceptable in that they rule out some plausible partial matches. A partial matching program, REIAX, is described which justifies the conclusion that a data element D is an unacceptable match to a pattern element P on the grounds that some other data element Db is a better match to P. The use of such relative criteria makes it possible to compute all of the kinds of mappings that Kling [9] has claimed are required to account for instutions about similarity. The ability to form unrestrictive mappings allows RELAX to construct generalizations that have eluded other approaches [5, 6].