Empirical comparison of two class model normalization techniques: Obstacles and questions

Author(s): J.R. Falleri, M. Huchard, C. Nebut
Venue: Empirical Studies of Model-Driven Engineering (ESMDE '08)
Date: 2008

Type of Experiement: Case Study

Quality
3

Summary:
This paper discusses the challenges of creating accurate models in the context of a generalizing model transformation. The authors do this by comparing, through an experiment, two approaches for improving model abstractions through improved generalizations. The two approaches are Formal Concept Analysis (FCA) and Relational Concept Analysis (RCA). FCA is a process of discovering accurate model generalizations. RCA builds off of FCA by providing more accurate generalizations, however many of the generalization produced by RCA may not be relevant.

The experiment tries to answer the following question: "Comparing generalizations produced by RCA versus those produced by FCA, and considering the effort needed for parametrizing and using results of FCA/RCA, is RCA an interesting improvement in practice?”

The results of the study demonstrated that RCA did in fact generate an excessive amount of irrelevant abstractions. In one of the experiments 1534 new classes were created by RCA. Interestingly the study didn't supply the results for FCA. Another experiment showed that RCA produced more inaccurate new concepts while FCA produced more inaccurate mergers. This isn't surprising because RCA produced many more concepts then FCA.

0