Mining human factors general trends from +100k UML class diagrams.

Models are primary artifacts for Model-Based Software Engineering. An important share of the activity associated with this engineering is creating and editing models (usually referred to as modeling). Using (graphical) diagrams to manipulate these models is one of the most common practices. A lot of scientific work has been performed on human factors related issues associated to these diagrams and their associated tools in order to support better the modeling activity through diagrams (referred to as diagramming [9]). However, our knowledge of the real practices concerning this modeling activity remains limited, especially when it comes to going into the detailed activity performed on a given type of diagram. In this article, we propose a first quantitative and detailed study of diagramming practices, focusing on UML class diagrams. Based on the observation of syntactic data of more than 100,000 UML class diagrams, we provide a list of general trends extracted from recurrent practices related to this type of diagram. These trends include the type of elements used, their frequency, the formatting of their names, their placement, and a possible coloring. These trends provide a sketch of typical construction and manipulation of a class diagram and provide input for editors to adapt their tools and for researchers to map practices and determine which aspects need to be studied further.