Résumé
The rapidly changing needs among other things due to technical innovation, competition and regulation often leads to describe the context for the study of conceptual models in information systems to facilitate the evolution of operating systems. The development of these models is carried out in several phases during which several working teams of different nature, providing each participant’s perception of the system to be built is limited to the part of his area of specialization. It must then reconcile the different perceptions.The main objective of the thesis is to design mechanisms to obtain a share of the model factoring concepts common to several models and, secondly, to provide designers with a methodology for monitoring the evolution of factorization.To perform the factorization, we have implemented the Formal Concept Analysis (FCA) and Relational Concepts Analysis (RCA), which are methods of analysis based on the theory of lattice data. In a set of entities described by features, both methods extract formal concepts that combine a maximumof entities to a maximumset of shared characteristics together. These formal concepts are structured in a partial order of specialization that provides with a lattice structure. The RCA can complement the description of the entities by relationships between entities.The first contribution of the thesis is a method a model for analyzing the evolution of the factorizationbased on the FCA and the RCA. This method builds the capacity of the AFC and the RCA to emerge in a model of thematic abstractions higher level, improving semantic models. We show that these methods can also be used to monitor the analytical process with stakeholders. We introduce metrics on the design elements and the concept lattices which are the basis for the development of recommendations. We conduct an experiment in which we study the evolution of the 15 versions of the model class of information-Pesticides EIS system.The second contribution of this thesis is a depth study of the behavior of the RCA on UML models. We show the influence of model structure on different variables studied (such as execution time and memory used) through several experiments on 15 versions of the EIS-Pesticides model. For this, we study several configurations (choice of elements and relations in the meta-model) and several parameters (choice of using unnamed elements, choice of using airworthiness). Metrics are introduced to guide the designer in managing the process of factoring and recommendations on the preferred configurations and settings are made.The last contribution is a approach to inter-model factorization to group in a model all the concepts common to different source models designed by different experts. In addition to the consolidation of common concepts, this analysis produces new abstractions generalizing existing thematic concepts. We apply our approach on 15 versions of the model EIS-Pesticides. All this work is part of a research framework which aims to factor thematic concepts within a model and control metrics by the profusion of concepts produced by the FCA and especially by RCA.