Recommending Auto-completions for Software Modeling Activities

Abstract

Abstract. Auto-completion of textual inputs benefits software developers using IDEs and editors. However, graphical modeling tools used to design software do not provide this functionality. The challenges of recommending auto-completions for graphical modeling activities are largely unexplored. Recommending auto-completions during modeling requires detecting meaningful partly completed activities, tolerating variance in user actions, and determining the most relevant activity that a user wants to perform. This paper proposes an approach that works in the background while a developer is creating or evolving a model and handles all these challenges. Editing operations are analyzed and matched to a predefined but extensible catalog of common modeling activities for structural UML models. In this paper we solely focus on determining recommendations rather than automatically completing an activity. We demonstrated the quality of recommendations generated by our approach in a controlled experiment with 16 students evolving models. We recommended 88% of the activities that a user wanted to perform within a short list of ten recommendations.

Pattern-based auto-completion of UML modeling activities

“…Furthermore, it allows a user to preview and to adjust a selected completion graphically before finally executing the completion. Our approach is based on the recognition of modeling activities and the computation of relevant recommendations, which we demonstrated in our previous work [8]. This approach operates on a set of predefined activity patterns.…”

“…We previously identified and iteratively evolved a comprehensive set of such modeling activities [8,9,10]. Examples of these activities are "replacing an association by an interface realization", "extracting an attribute into a class", or "replacing an association class".…”

“…In previous work, we showed that partly performed modeling activities can be recognized and suggested for autocompletion while a developer is creating or evolving UML models [8]. The proposed technology captures editing operations within a modeling environment and matches them to predefined activity patterns.…”

Recommender systems in model-driven engineering

“…For example, SimVMA [132] provides step-wise suggestions to evolve Simulink models based on model clone analysis; Heinemann [49] recommends elements defined in model libraries (e.g. blocks from Simulink libraries) based on data mining existing models; the approach by Kögel et al [68,69] recommends model changes applicable to the same context of the last model change; DoMoRe [5,6] suggests domain concepts and names for new model elements; RapMOD [72,73] offers auto-completion actions for (UML) graphical models, similarly to the vision paper [121]; Elkamel et al [37] recommend UML classes that are similar to the ones in the UML class diagram being developed; Li et al [75] and Deng et al [31] recommend activity nodes for process models; Rangiha et al [110] recommend tasks and actor roles in a social process-modelling tool; Koschmider et al [50,51,71] recommend process fragments to complete a process model; Baya [27] recommends mashup model patterns based on the context, the user and different expert recommendations, and helps in weaving the selected pattern into the partial model under development; and Hermes [34][35][36] permits building Eclipse-based RSs that help in completing models with recommended elements from other models in a repository. Instead of profiting from repositories of models, PME's recommendations are based on an analysis of the language meta-model [99].…”

A definition-by-example approach and visual language for activity patterns in engineering disciplines

“…In previous work, we used complex patterns of editing events to update traceability information automatically upon recognizing users' modeling activities [7] and successfully brought a visual language approach to traceability analysis [8][9][10]. In a succeeding project, we studied how software modeling activities can be recommended while a developer is editing UML models [11] and how such recommended activities can then be auto-completed [12][13][14]. Both proposed approaches build upon activity definitions captured as event-processing rules.…”