Improving Model Repair through Experience Sharing.

Abstract

In model-driven software engineering, models are used in all phases of the development process. These models may get broken due to various editions throughout their life-cycle. There are already approaches that provide an automatic repair of models, however, the same issues might not have the same solutions in all contexts due to different user preferences and business policies. Personalization would enhance the usability of automatic repairs in different contexts, and by reusing the experience from previous repairs we would avoid duplicated calculations when facing similar issues. By using reinforcement learning we have achieved the repair of broken models allowing both automation and personalization of results. In this paper, we propose transfer learning to reuse the experience learned from each model repair. We have validated our approach by repairing models using different sets of personalization preferences and studying how the repair time improved when reusing the experience from each repair.

Recommender systems in model-driven engineering

“…These approaches differ either in the applied technique to compute and rank the repairs, or in the application domain. In particular, IntellEdit [93] ranks quick fix solutions to model inconsistency problems according to the least-change principle; PAR-MOREL [11,53] determines the model repair actions based on the user preferences and on the experience gained from repairing under different personalisation settings; the diagram predicate framework (DPF) [108] and the approach by Nassar et al [91] implement repairs as transformation rules; DiaGen [82] represents models as hypergraphs and uses hypergraph patches to produce recommendations for repairing models; Refacola [130] uses constraint-based rules; BPMoQualAssess [60] provides guidelines to improve the actual value of quality metrics for business process models; B-repair [24] is specific to the B formal specification language and ranks the suggested repairs based on their estimated quality; Revision [98] tracks model inconsistencies to the editing action originating them in the model history and fixes this action to obtain a consistent model; MDSafe-Cer [87] detects missing information for supporting key evidence in process-based argumentations, and recommends how to resolve such deviations; ASIMOV [38] assists in the co-evolution of models and meta-models by proposing model co-evolution actions that a metamodeller must have defined previously; and Anguel et al [8] also tackle the co-evolution problem, but they automatically fix resolvable changes and recommend coevolution actions to deal with non-resolvable changes.…”

“…In addition, two of the model repair approaches can be used to repair meta-models as well. PARMOREL allows repairing meta-models having duplicate attributes in related classes, or properties modelled both as attributes and as references [11]. Refacola [130], on the other hand, can help repairing syntactically incorrect meta-models, e.g.…”

“…For this purpose, they are frequently defined over a meta-modelling frameworklike the Eclipse Modeling Framework (EMF) [131], Kermeta [57], GME [74], or DPF [109]-which enables their application to models of any language defined within the framework. This is the case of [11,[34][35][36]53,68,69,86,91,93,98,99,108,[123][124][125][126][127]130]. In the case of DiaGen [82], language independence is achieved by representing models as hypergraphs, and language definitions as hypergraph grammars.…”

“…In the case of DiaGen [82], language independence is achieved by representing models as hypergraphs, and language definitions as hypergraph grammars. Other approaches are meta-level independent, and since metamodels are also models, such approaches are suitable for both meta-models and models [5,6,11,[34][35][36]53,68,69,[123][124][125]130]; however, when applied to meta-models, the approaches are dependent on the meta-modelling language used, like EMF's Ecore. Finally, language independence can be a gradual term.…”

Multi-Model Evolution through Model Repair.

“…Learning, Incrementality and Invocation "Learning is the ability of a program to improve its performance on a given task over time" (Russell & Norvig 2010). It appears to be promising enhancement for improving the performance of search-based approaches (Barriga et al 2018) and can help to identify hidden policies and user preferences (Barriga et al 2020;Ludovico et al 2020).…”