Abstract The application of Large Language Models (LLMs) in Model-Driven Engineering (MDE) has emerged as a rapidly evolving research area. While existing systematic literature reviews have examined specific technical approaches, a comprehensive mapping of the broader research landscape (e.g., development trends) remains lacking. This study presents a systematic mapping study of LLM applications in MDE, analyzing 86 primary studies collected from five databases, covering publications from 2022 to early 2026. Guided by five research questions, we characterize the field across five dimensions: MDE task distribution and research contribution types, LLM technologies and interaction strategies, artifact representation and processing, validation practices, and publication landscape. Our findings reveal that current LLM4MDE research is heavily concentrated on Model Generation, while tasks such as Model Migration, DSL Engineering, and Metamodeling remain marginal. Most approaches rely on black-box OpenAI models accessed via remote APIs and adapted through prompt engineering, with fine-tuning and retrieval-augmented generation rarely employed. Inputs are predominantly natural-language artifacts, while outputs are model-oriented but usually expressed in lightweight textual formats rather than native MDE exchange formats. Validation is centered on quantitative experimentation, with 42% of studies reporting no baseline and cost efficiency reported in fewer than one quarter of studies. The field has grown rapidly, from one paper in 2022 to 42 in 2025, with research concentrated in Europe and Canada and limited industry involvement. Based on these findings, we identify gaps and opportunities across task coverage, technical configuration, and evaluation practice, offering a knowledge map to guide future work in this cross-disciplinary field.

Ensuring the logical coherence of subsumption hierarchies is essential for high-quality ontology engineering. While validation frameworks such as Onto-Clean provide formal mechanisms for checking taxonomic consistency in endurant entities, there remains a critical gap in validation tools for perduring entities such as processes, events, and states. This paper introduces a principled framework for validating perduring entity subsumption hierarchies by leveraging meta-properties, including cumulativity, homeomericity, atomicity, and agentivity. The framework defines a set of meta-property constraints that enforce consistency in the subsumption hierarchy of perdurants. Case studies demonstrate its utility across both foundational ontologies and domain-specific levels. Notably, when applied to the uppertier structure of DOLCE, the framework preserves its internal stratification while exposing classification gaps. Furthermore, applying the framework to the MAVEN event-type hierarchy reveals ontological inconsistencies introduced by human developers, demonstrating the framework’s effectiveness as a verification tool. This work establishes a foundation for the meta-property-driven construction and validation of perdurant taxonomies, addressing a long-standing gap in formal ontology validation methodologies.

The dependency on extensive expert knowledge for defining subgoals in hierarchical reinforcement learning (HRL) restricts the training efficiency and adaptability of HRL agents in complex, dynamic environments. Inspired by human-guided causal discovery skills, we proposed a novel method, Human Causal Perception and Inference-driven Hierarchical Reinforcement Learning (HCPI-HRL), designed to infer diverse, effective subgoal structures as intrinsic rewards and incorporate critical objects from dynamic environmental states using stable causal relationships. The HCPI-HRL method is supposed to guide an agent's exploration direction and promote the reuse of learned subgoal structures across different tasks. Our designed HCPI-HRL comprises two levels: the top level operates as a meta controller, assigning subgoals discovered based on human-driven causal critical object perception and causal structure inference; the bottom level employs the Proximal Policy Optimisation (PPO) algorithm to accomplish the assigned subgoals. Experiments conducted across discrete and continuous control environments demonstrated that HCPI-HRL outperforms benchmark methods such as hierarchical and adjacency PPO in terms of training efficiency, exploration capability, and transferability. Our research extends the potential of HRL methods incorporating human-guided causal modelling to infer the effective relationships across subgoals, enhancing the agent's capability to learn efficient policies in dynamic environments with sparse reward signals.

Achieving real-time agility and adaptation with respect to changing requirements in existing IT infrastructure can pose a complex challenge. We describe a goal-oriented approach to manage this complexity. We argue that a goal-oriented perspective can form an effective basis for devising and deploying responses to changed requirements at runtime. We offer an extended vocabulary of goal types by presenting two novel conceptions: differential goals and integral goals, which we formalize in both linear-time and branching-time settings. We describe goal lifecycles and interactions and the extended notion of context for the representation of rapidly changing, complex operating environments. We then illustrate the working of the approach by presenting a detailed scenario of adaptation in a Kubernetes setting, in the face of a Distributed Denial-of-Service (DDoS) attack.

The integrated management of industrial systems in future environments like Industry 4.0 requires the effective management of information throughout the engineering life cycle. As systems pass through phases of design, construction, operation, maintenance, renewal or replacement, they will be administered via different information ecosystems, requiring changing perspectives on their descriptive information. A central role in the interplay of software and hardware artefacts, functions, documentation and managing software is played by the descriptions of concepts (i.e. formalised definitions of concepts within the domain of quantification). In this paper we propose a unified formalisation of descriptions that permits consistent analysis of the relationships between the designs, types, products, and concrete artefacts that can be found in the industrial engineering life-cycle. The approach is consistent with our earlier framework that describes artefacts, requirements and functional roles in the context of the DOLCE foundational ontology.

Traditionally the integration of data from multiple sources is done on an ad-hoc basis for each analysis scenario and application. This is an approach that is inflexible, incurs high costs, and leads to “silos” that prevent sharing data across different agencies or tasks. A standard approach to tackling this problem is to design a common ontology and to construct source descriptions which specify mappings between the sources and the ontology. Modeling the semantics of data manually requires huge human cost and expertise, making an automatic method of semantic modeling desired. Automatic semantic model has been gaining attention in data integration [5], federated data query [14] and knowledge graph construction [6]. This paper proposes an service-oriented architecture to create a correct semantic model, including annotating training data, training the machine learning model, and predict an accurate semantic model for new data source. Moreover, a holistic process for automatic semantic modeling is presented. By the usage of ASMaaS, historical semantic annotations for training machine learning model used in automatic semantic modeling can be shared, reducing costs of human resources from users. By specifying a well defined interface, users are able to have access to automatic semantic modeling process at any time, from anywhere. In addition, users must not be concerned with machine learning technologies and pipeline used in automatic semantic modeling, focusing mainly on the business itself.

The following topics are dealt with: business data processing; ontologies (artificial intelligence); data mining; security of data; cryptography; cloud computing; software architecture; query processing; storage management; distributed databases.

Entity resolution (ER) is the problem of accurately identifying multiple, differing, and possibly contradicting representations of unique real-world entities in data. It is a challenging and fundamental task in data cleansing and data integration. In this work, we propose graph differential dependencies (GDDs) as an extension of the recently developed graph entity dependencies (which are formal constraints for graph data) to enable approximate matching of values. Furthermore, we investigate a special discovery of GDDs for ER by designing an algorithm for generating a non-redundant set of GDDs in labelled data. Then, we develop an effective ER technique, Certus, that employs the learned GDDs for improving the accuracy of ER results. We perform extensive empirical evaluation of our proposals on five real-world ER benchmark datasets and a proprietary database to test their effectiveness and efficiency. The results from the experiments show the discovery algorithm and Certus are efficient; and more importantly, GDDs significantly improve the precision of ER without considerable trade-off of recall.

In the era of big data, new challenges occur in the field of data federation research. New types of data sources with new formats of data have emerged, and end users need to conduct complex search and data analytical tasks, which impose requirements such flexible data federation, customized security mechanism and high-performance processing (for example, near real time query). To address these challenges, this paper proposes a data federation platform named FEDSA and reports on an initial implementation. Distinctive features of the platform include process-driven data federation, Data Federation as a Service, a simple query language over a high-level common data model, data security protection over all federation services, query re-writing and full distribution. We demonstrate how these features address the challenges, discuss the performance of the current implementation, and outline future extensions.

Due to the volume, variety, and veracity of network data available, information fusion and reasoning techniques are needed to support network analysts’ cyber-situational awareness. These techniques rely on formal knowledge representation to define the network semantics with data provenance at various levels of granularity. To this end, this paper proposes the Communication Network Topology and Forwarding Ontology, a state-of-the-art ontology that enables the formal, unified representation of complex network concepts regardless of the type of the data source. The implementation of this ontology allows network analysts to represent expert knowledge and query network data fused from disparate data sources.

Model transformations are an important aspect of Model-Driven Engineering as models throughout the software development process are transformed and refined until, finally, application code is generated. However, model transformations are complex to build, maintain, and verify for correctness. We propose the combination of visual contracts, an implementation independent approach for specifying correctness requirements for verification purposes, with operator-based model transformation execution to integrate both the specification of transformation requirements and the transformations themselves within the same framework. The graphical operator-based notation is used to define both the constraints of a contract and the transformation definition. This allows reuse of operators between the two and maintains implementation independence as the operators can be directly executed or compiled to other model-transformation languages. To illustrate the concept, we report on a prototype integration of visual contracts with our our existing operator-based model transformation framework and applied it in the law-enforcement context to transform relational data sources into Elasticsearch for multisource analytics.

In large organizations, multiple stakeholders may modify the same business process. This paper addresses the problem when stakeholders perform changes on process views which become inconsistent with the business process and other views. Related work addressing this problem is based on execution trace analysis which is performed in a post-analysis phase and can be complex when dealing with large business process models. In this paper, we propose a design-based approach that can efficiently check consistency criteria and propagate changes on-the-fly from a process view to its reference process and related process views. The technique is based on consistent specialization of business processes and supports the control flow aspect of processes. Consistency checks can be performed during the design time by checking simple rules which support an efficient change propagation between views and reference process.

Effective exchange of information about processes and industrial plants, their design, construction, operation, and maintenance requires sophisticated information modelling and exchange mechanisms that enable the transfer of semantically meaningful information between a vast pool of heterogeneous information systems. In order to represent entities relevant to the engineering life-cycle, social concepts, descriptions, roles, artefacts, functions, and information objects must be integrated in a coherent whole. Forming the basis of this integration in our framework is the DOLCE foundational ontology. In this paper we propose an ontologically well-founded approach to modelling artefacts, their requirement specifications and functional roles, such that consistency of their relationships in the data model can be verified. Specifically, we discuss the modelling of engineering artefacts, roles and role-filling capacity in the context of data modelling for industrial information exchange.

This paper addresses the problem of transforming natural language business specifications into formal models suitable for use in information systems. In particular, the Semantics of Business Vocabulary and Business Rules (SBVR) standard is used as a starting point for both the natural language specifications and the formal representation. In recent years, SBVR-based approaches have been proposed for transforming natural language business rules into models, such as UML Class Diagrams, however, most focus on the transformations performed after the SBVR models have been created and, therefore, simplify or entirely neglect the natural language to SBVR transformation. There are a number of difficulties in transforming natural language into formal models, such as ambiguity and inconsistency. This paper presents a solution based on a unique combination of techniques from Cognitive Linguistics and Knowledge-based Configuration in order to transform natural language business specifications into SBVR models. We present a comparative survey of state-of-the-art approaches and argue that current solutions do not fulfil the criteria necessary to meet our goals. We demonstrate our approach and show how it improves translation of natural language business specifications into formal models and increases the level of automation for Model-Driven Engineering.

The ability to build new (complex) services by composing existing services is one of the key benefits of the Service Oriented Architecture paradigm. Existing approaches to automate composition requires pre-planning or prediction of the number of required services, making them unsuitable in dynamic composition scenarios. To address this gap, we present a consistency-based service composition approach, where composition problems are modeled in a generative constraint-based formalism. We illustrate how the configuration of service processes differs from established constraint-based configuration techniques and develop an algorithm to synthesis valid service process compositions. We also show that our technique scales well to non-trivial problems.

Automated fault diagnosis has significant practical impact by improving reliability and facilitating maintenance of systems [1]. Given a monitor continuously receiving observations from a dynamic eventdriven system, diagnosis algorithms infer possible fault events that

Developing model-based automatic debugging strategies has been an active research area for several years, with the aim of locating defects in a program by utilising fully automated generation of a model of the program from its source code. We provide an overview of current techniques in model-based debugging and assess strengths and weaknesses of the individual approaches. An empirical comparison is presented that investigates the relative accuracy of different models on a set of test programs and fault assumptions, showing that our abstract interpretation based model provides high accuracy at significantly less computational effort than slightly more accurate techniques. We compare a range of model-based debugging techniques with other state-of-the-art automated debugging approaches and outline possible future developments in automatic debugging using model-based reasoning as the central unifying component in a comprehensive framework.

Abstract:As ontologies become more prevalent for information management the need to manage the ontologies increases. Multiple organizations, within a domain, often combine to work on specific projects. When separate organizations come together to communicate, an alignment of terminology and semantics is required. Ontology creation is often privatized for these individual organizations to represent their view of the domain. This creates problems with alignment and integration, making it necessary to consider how much each ontology should influence the current decision to be made. To assist with determining influence a trust‐based approach on authors and their ontologies provides a mechanism for ranking reasoning results. A representation of authors and the individual resources they provide for the merged ontology becomes necessary. The authors are then weighted by trust and trust for the resources they provide the ontology is calculated. This is then used to assist the integration process allowing for an evolutionary trust model to calculate the level of credibility of resources. Once the integration is complete semantic agreement between ontologies allows for the revision of the authors' trust.

Modelling complex processes in the service oriented architecture paradigm typically requires the composition of a number of simpler services to achieve a desired goal. We present a semantic service composition approach that is based on the use of UML activity diagrams as abstract specification language and propose a service selection process that combines both conceptual and instance-level analysis to locate suitable services. Model-driven principles allow to adhere to design requirements that would otherwise be difficult to incorporate informal composition frameworks. Explicit modelling of behaviour in the presence of failures permit to create workflows that are robust in case of service execution errors. This approach may allow to synthesise concrete service orchestrations that are superior to similar approaches purely based on type-based or conceptual matchmaking.

The ISO-10303 standard, called STEP, originated in 1984 and published in 1994, has been adopted by many industries including shipbuilding, aerospace and automotive. This paper throws new light on the well established standard by investigating the suitability of STEP to support the exchange of parameterised models between CAD systems. The paper points out how design intent, i.e. parameters, constraints and constructional history of a model, can be transferred to another system and it explains why the new enhancements of STEP which were published recently are able to ease the migration of a STEP-based model. The implications of these enhancements on the integration of the automotive design environment are outlined shortly and provide a showcase of these refinements.

In this paper, we describe use of constraint-based methods for configuring ubiquitous workspaces. A declarative representation allows succinct, easily maintainable definitions of the dependencies inherent in setting up a meeting, and permits the use of general constraint reasoners for various standard tasks such as setting up meeting interfaces, switching between setting for different meetings, and saving and restoring settings. Personalisation techniques can be used for intelligently adapting the workspace to individual user needs.

Even with modern software development methodologies, the actual debugging of source code, i.e., location and identification of errors in the program when errant behavior is encountered during testing, remains a crucial part of software development. To apply model-based diagnosis techniques which have long been state of the art in hardware diagnosis, for automatic debugging a model of a given program must be automatically created from the source code. This work describes a model that reflects the sequential execution semantics of the Java language, including exceptions and unstructured control flow, thereby providing unprecedented scope in the application of model-based diagnosis to programs. Notably, this approach omits the strict view of a component representing one statement of earlier work and provides a more flexible mapping from code to model.

Object-oriented design methodologies represent the behavior of instances of an object type not merely by a set of operations, but also by providing an overall description on how instances evolve over time. Such a description is often referred to as "object life cycle."Object-oriented systems organize object types in hierarchies in which subtypes inherit and specialize the structure and behavior of their supertypes. Past experience has shown that unrestricted use of inheritance mechanisms leads to system architectures that are hard to understand and to maintain, since arbitrary differences between supertype and subtype are possible. Evidently, this is not a desirable state of affairs and the behavior of a subtype should specialize the behavior of its supertype according to some clearly defined consistency criteria. Such criteria have been formulated in terms of type systems for semantic data models and object-oriented programming languages. But corresponding criteria for the specialization of object life cycles have so far not been thoroughly investigated.This paper defines such criteria in the realm of Object Behavior Diagrams, which have been originally developed for the design of object-oriented databases. Its main contributions are necessary and sufficient rules for checking behavior consistency between object life cycles of object types in specialization hierarchies with multiple inheritance.

Model-based diagnosis is a successjid AI technique for locating and identifying faults in technical syslems. Extending previous research on model-based diagnosis support for fault search in technical designs, we are building a model-based debugger for Java programs to provide intelligent support for the programmer trying to locate the so~rce of an error: By using one or more models derived from the source code of the program without additional spec$cations except the Java semantics, the debugger guides the user towards potential sources for incorrect program behaviors, i.e., bugs.

The state of the art in hardware design is the use of hardware description languages such as VHDL. The designs are tested by simulating them and comparing their output to that prescribed by the specification. A significant part of the design effort is spent on detecting unacceptable deviations from this specification and subsequently localizing the sources of such faults. In this paper, we describe an approach to employ model-based diagnosis for fault detection and localization in very large VHDL programs, by automatically generating the diagnosis model from the VHDL code and using observations about the program behavior to derive possible fault locations from the model. In order to achieve sufficient performance for practical applicability, we have developed a representation that provides a highly abstracted view of programs and faults, but is sufficiently detailed to yield substantial reductions in the fault localization costs when compared to the current manpower-intensive approach. The implementation in conjunction with the knowledge representation is designed with openness in mind in order to facilitate use of the highly optimized simulation tools available.

To model organizational changes in cooperative but distributed problem solving environments, it is necessary to represent not only activities and knowledge concerning particular agents, but also the constraints for successful collaboration among agents, concerning mainly the availability of common knowledge and cooperative task accomplishment. The usability of such models mostly depends on their ability to describe communication events occurring in multiagent environments. To this end, the authors enhance a formal communication model with adaptation capabilities to capture organizational changes in cooperative problem solving environments. The model meets the appropriate level of abstraction not only for the specification of the communication among a dispersion set of agents, but also for concurrent common, but distributed knowledge, and related (a)synchronous events.>