Similarity Measure and Mapping Methods for Comparing Object ModelsAt Different Abstract Levels

摘要

In this paper, we present a method for comparing modelsrnat different abstract levels. Because of the lack ofrninformation about implementation in the phase ofrnrequirement analysis, the produced object model is morernabstract than the extracted implementation model.rnTherefore, the automatically extracted model is morernconcrete and less abstract than the domain object model.rnTo solve this discrepancy, we use the word matchingrntechnique for comparing the two models in the samerndomain in our method. Class consists of many identifiersrnsuch as a class name; attribute names and method names.rnIdentifiers can be divided into separated words with therntechniques used for the identifiers. Then, we find a class ofrna model that matches with the class of another modelrnthrough calculating the number of matched words. In otherrnwords, our method finds a class in one model that matchesrnmost to a class in the other model.rnAfter the above match step, we can distinguish structuralrnsimilarities between two models, because there existrnrelations between the entities. Comparing the two models,rnthe number of classes increases in implementation modelrnand the contents of a class of implementation model arernadded and changed. With our method, reengineers canrnobtain information on structural similarities informationrnusing the method that overlaps produced object model andrnautomatically extracted object model on a screen.rnThe method we propose is being incorporated in arnproject called REsearch on object-oriented SOftwarernReengineering Technology(RESORT) which develops arnsoftware-reengineering tool.