Towards Machine Learning for Learnability of MDD Tools

Saad bin Abid, Vishal Mahajan und Levi Lúcio

Software Engineering and Knowledge Engineering (SEKE) Conference, Lisbon, Portugal, pp. 1–6

Juli 2019 · DOI: 10.18293/SEKE2019-050


Learning how to build software systems usingnew tools can be a daunting task to anyone new to the job.This is especially true of tools that provide a large numberof functionalities and views on the system under development,such as IDESfor Model-Driven Development (MDD). ApplyingMachine Learning (ML) techniques can help in this state ofaffairs by pointing out to appropriate next actions to rookieor even intermediate developers. AutoFOCUS3 (AF3) is amature MDD tool we are building in-house and for which weprovide regular tutorials to new users. These users come fromboth the academia (e.g, students/professors) and the industry(e.g. managers/software engineers). Nonetheless, AF3 remainsa complex tool and we have found there is a need to speedupthe learning curve of the tool for students that attend ourtutorials – or alternatively and more importantly for others thatsimply download the tool and attempt using it without humansupervision. In this paper, we describe a machine learning-basedrecommendation system named MAGNETfor aiding beginnerand intermediate users of AF3 in learning the tool. We describehow we have gathered data and trained an ML model to suggestnew commands, how a recommender system was integrated inthe AF3, experiments we have run thus far, and the futuredirections of our work.

Stichworte: Model-Driven Development, MDD, AutoFOCUS3, Machine Learning, Intelligent Recommendation Systems, IRS, Eclipse IDE, Domain-Specific Languages, DSLs, development interaction data, methodology, tooling, model-based systems engineering, MbSE