In architecture, the construction process underwent a major shift in the last decade with the increasing adoption of Building Information Modeling (BIM) processes to design and handle the construction cycle. Starting from a 3D model containing the information relative to an architectural project, the BIM objective is to supervise the entire construction project, from the crane positioning to the interior design with lighting simulation and rendering pictures. The BIM have been mostly studied from a technical perspective \cite{Wang_2012,Ernstrom2006,Kymmell2008} with the common assumption that different user profiles, with complementary area of expertise, contribute together to BIM models. In practice, as each user adds his own objects, attach his personal annotations, the resulting model gathers a very large amount of information, from various departments, with various level of details and regarding different moment in the building lifetime (from construction to exploitation).
BIM processes are powerful conceptual tools, nonetheless several limitations have restricted the integration in the industrial world. The multiple difficulties to implement BIM processes in construction firms are documented in \cite{Smith2013}. Coping with the industrial reality, BIM models display huge variations depending of their authors, the local conception methods and the contained words in various languages. Without effective standard directives, all the produced models are complex, hard to manage for any specialist, and the resulting BIM processes fail to unify the conception methods. This discrepancy represents a important lack of productivity for the construction industry.
In this paper, we propose a method to create an adaptive environment built on machine learning and visual saliency approaches to properly take into consideration the user's profile. The contributions described in this paper are the following:
The organization of the paper is the following. In Section \ref{sec:soa}, we review the current trends in BIM, its limitations and the saliency, a central notion linked to produce efficient visualization. In Section XXX, we propose a model to classify automatically the BIM elements, with respect to their construction departments, as shown on the central part of Figure \ref{837307}. The proposed method is evaluated on BIM data, with experiments taking into account the wide diversity of teams working on a single construction model. Section XXX introduces an algorithm to computes saliencies adapted to architectural constraints. A system is proposed to enhanced the visualization of BIM models by integrating information relative to the user's department, an example is shown on the right part of Figure \ref{837307}.  This system is evaluated by industrial building experts. Section XXX concludes this paper.