Marcos Carrasco Sacristán, Óscar Pastor López, Ana León Palacio and Juan Carlos Casamayor Rodenas
The rapid growth of omics resources has led to a highly fragmented and heterogeneous landscape, giving rise to what we refer to as “genomic chaos”, in which data, information, and knowledge are represented at different levels of abstraction across genomic, transcriptomic, proteomic, reaction, and pathway domains. This heterogene- ity, together with the lack of consistent cross-domain standardisation, complicates semantic interoperability, data integration, and the development of bioinformatics systems. In this context, we present the MOSAIC model (Multi-Omics Structured and Integrated Conceptual model), an extensible conceptual and logical reference model that structures core omics entities, their descriptions, and their relationships into an information architecture. The model is designed to capture how variants, genes, transcripts, proteins, diseases, reactions, pathways, functional roles, and chemical context relate across omics domains, providing a coherent basis for representing domain knowledge in a structured way. Rather than defining a specific physical implementation, the model establishes a reusable logical foundation for aligning and integrating knowledge from complementary domain resources, standards, and future bioinformatics applica- tions. By addressing the “genomic chaos” generated by inconsistent representations and fragmented repositories, this work proposes a versioned and extensible model for organising multi-omics knowledge, whose scope can evolve as curated resources and scientific evidence introduce relevant biological distinctions.

Marta Vivó-Benlloch, Mireia Costa, Ana Pérez-Gómez, Alberto García S., Ana León, José Fabián Reyes Román, Eloisa Jantus-Lewintre, Sandra Gallach and Óscar Pastor
Artificial intelligence (AI) holds significant promise for supporting complex medical decisions, but the opacity of high-performing models hinders its integration into clinical practice. Within the field of Explainable AI (XAI), post-hoc explainability techniques such as SHAP and LIME have emerged as solutions for making model behavior more transparent. However, their ability to support clinically meaningful explanations is limited by the semantic quality of the variables they describe. When clinical data are heterogeneous, loosely defined, or semantically inconsistent, post-hoc explanations may be technically valid but clinically meaningless. This paper argues that conceptual modeling provides the semantic layer needed to overcome this limitation and reports its application in the domain of liquid biopsy through a research collaboration with a Hospital Research Foundation. The main contribution is a formal domain representation comprising a conceptual model and a glossary of terms that standardize the domain’s concepts, make their interrelationships explicit, and address the heterogeneity inherent in raw clinical data. In doing so, it establishes a semantically grounded foundation on which AI algorithms can be built, evaluated, and explained meaningfully. The proposed representation has been validated through focus groups with domain experts, and its usability has been further evaluated with medical experts and non-expert students from biomedical and informatics engineering.

Elena Milivinti
Enforcement authorities have identified a particular documentary combination as a direct enabler of illicit waste trafficking: the same physical consignment framed simultaneously as traded goods under one classification system and as waste under another. This paper addresses that problem through the case of textile waste, where the European Waste Catalogue (EWC) classifies consignments by generating process while the Harmonized System (HS), developed by the World Customs Organization, classifies by material composition or use history. The analysis covers eight textile-relevant EWC codes and twelve textile-relevant HS headings, generating 96 mapping pairs. Using Basic Formal Ontology as an analytical framework, the paper makes explicit the properties each code requires, and evaluates each pair according to whether the source code already satisfies the definitional requirements of the target code, whether additional evidence is needed, or whether translation is blocked. No pair qualifies as an exact match. A documented textile shipment shows how the method identifies apparently routine documentary combinations that require review by a competent authority. The paper contributes a conceptual-modeling method for validating translations between incompatible classification systems and for making classification gaps explicit, auditable, and usable in digital waste-tracking systems.

Zixin Wu and Janis Voigtländer
Conceptual modeling education requires large numbers of tasks that are structurally valid, didactically useful, and sufficiently diverse for repeated practice. Existing approaches, however, expose a clear trade-off. Formal-methods-based generation provides controllability, validation, and explicit difficulty configuration, but usually produces tasks with artificial identifiers and weak contextualization. Large language models (LLMs), in contrast, can provide rich real-world semantics, but lack reliable formal guarantees and fine-grained control. To address this tension, we propose FoLIC (formal-constraint-grounded LLM-assisted semantic injection and contextualization), a decoupled framework that integrates LLM-based semantic contextualization into formally grounded conceptual-modeling task generation. FoLIC preserves formal validity by separating solver-backed structural generation from semantic enrichment, so that LLMs are confined to concept injection and scenario construction while formal constraints remain under explicit control. Our evaluation shows that FoLIC preserves formal validity by design, supports contextualized task variants across multiple task families, and en- ables exploratory deployment in educational practice. At the same time, the results indicate that semantic quality depends strongly on the capability of the underlying LLMs.

Sarmad Rezayat, Daniel Garcia Vasquez, Gerrit Burmester, Hui Ma and Sven Hartmann
Ontology engineering methodologies typically optimize for semantic completeness, formal expressiveness, and interoperability, which are properties geared toward human reasoning and system integration. When such ontologies are used as extraction schemas for Large Language Models (LLMs), and particularly for local LLMs deployable under the data privacy and security constraints of safety-critical domains, these properties are necessary but not sufficient. The ontology must also be reliably operationalizable by a constrained extractor. We argue that extractability should be treated as a first-class concern in conceptual modeling alongside the established ones. We introduce Extraction-Oriented Ontology Modeling (EOOM), a six-stage conceptual modeling approach that integrates ontology refinement into the Knowledge Graph (KG) construction process rather than treating the guiding ontology as a fixed precondition. We apply EOOM to an aviation safety use case based on 643 accident reports, where the 915-class Air Traffic Management (ATM) Information Reference Model Ontology (AIRM-O) fails to support extraction by local LLMs. In contrast, a 20-class extraction schema produced using EOOM enables stable extraction across six local LLMs with varying sizes and architectures. The results show that ontology design is a stronger determinant of extraction quality than model choice, and that extraction schemas are a distinct conceptual artifact from domain ontologies, with different design requirements.

Jonathan Silva Mercado, Qin Ma, Jordi Cabot, Pierre Kelsen and Henderik A. Proper
The use of Large Language Models (LLMs) has shown significant potential for generating domain models from textual descriptions. However, these models often fall short in soundness, which is inevitably caused by the statistical nature of LLMs that only generate plausible output rather than output grounded in the provided input. Generated unsound model elements lack supporting evidence in the source textual descriptions, and are commonly referred to as hallucinations.
In this paper, we propose an explainable approach that combines (1) attention-based model element confidence scoring and (2) confidence- driven model ensembling to reduce hallucinations in domain model gen- eration from text. The proposed scoring mechanism leverages the attention weights within transformer architectures to estimate the strength of original textual evidence supporting each generated model element, distinguishing between intra-element and inter-element evidence. The confidence-driven model ensembling strategy then retains the most confident model elements from multiple executions of multiple LLMs and synthesizes them into a final domain model.
We evaluate the effectiveness of our approach using open-weight Small Language Models (SLMs), as their higher propensity for hallucinations provides a rigorous environment for validation. Through an ablation study, we demonstrate how confidence filtering based on intra-element and inter-element scores, together with an ensemble strategy, balances textual grounding and model completeness, providing interpretable evidence for the resulting domain models.

Wolfgang Maass and Iris Reinhartz-Berger
Existing agent-oriented conceptual modeling frameworks assume components with deterministic execution semantics. Applied to language model agents, whose outputs arise from probabilistic text generation governed by natural-language inputs, they lack three constructs that accountability requires: a binary declaration of each agent’s legitimate action space; runtime enforcement of negative obligations by a component external to the agent; and treatment of inter-agent communication as auditable evidence.
To address these gaps, we introduce AI-Intent, a conceptual modeling framework organized into three pillars: Declaration encodes each agent’s action boundary, decision rights, capabilities, and risk policies in Mandates; Enforcement evaluates every Proposed Action against the applicable Mandate through a Compliance Agent before delivery; Auditability derives a structured, durable Accountability Trace from every session. A reference implementation for private investment advisory un- der MiFID II is evaluated across 190 evaluation sessions: boundary violation containment averaged 97.9% with zero forced-pass occurrences, while audit trace completeness was found to depend on the instruction-following capability of the deployed model.

Aisha Aijaz, Raghava Mutharaju and Manohar Kumar
Applied ethics applies ethical decision-making to domain- specific contexts using contextual information such as agents, actions, temporal and spatial settings, and theoretical constructs such as utility, virtues, rights, and duties. However, representing an ethical decision is challenging as it may be abstract, context-sensitive, and semantically heterogeneous. Nevertheless, important ethical and contextual factors can be formally modeled to support structured ethical reasoning. Knowledge representation and reasoning provide a mechanism to translate abstract ethical concepts into machine-interpretable conceptual structures in the context of an event. To achieve this, we propose ApplE, an Applied Ethics ontology that models ethical theory and event context within a unified and modular conceptual framework for ethical decision-making. The ontology was developed using a modified version of the Simplified Agile Methodology for Ontology Development (SAMOD), which facilitates iterative refinement of classes and relationships, as well as the participation of a domain expert. The modular development of ApplE combines Ethics Theory with Event Context to capture semantic relationships between ethical principles, agents, actions, consequences, in- tentions, and domains. Using ApplE, we modeled a use case from the medical domain to demonstrate the ontology’s representational expressivity and reasoning capabilities. In addition to ontological reasoning and consistency checks, ApplE is also evaluated using the three-fold testing process of SAMOD. ApplE follows the FAIR principles and is positioned to be used as a reusable semantic and conceptual modeling resource for ethical AI systems and ontology-driven applications.

Alberto García S., Claudia Negri-Ribalta, René Noel and Oscar Pastor
Information systems that support transborder personal data flows (TPDFs) must be specified against data protection regimes that partly overlap, partly diverge, and may apply simultaneously to the same processing configuration. This exposes a conceptual gap in requirements engineering: regulatory data protection concerns must be represented in a form that supports their translation into system specifications while preserving the distinction between jurisdiction-agnostic commitments and jurisdiction-specific legal variation. Existing data protection ontologies provide valuable support for compliance modeling, but they remain largely centered on privacy or on specific regulations such as the GDPR. This paper addresses this gap by proposing a jurisdiction-agnostic reference ontology for Regulatory Data Protection Requirements (RDPRs). The ontology is based on the Unified Foundational Ontology (UFO) and was created following SABiO 2.0, using a thematic synthesis of interviews with legal experts and an analysis of data protection regulations across several jurisdictions. It captures the common structure linking personal data processing, governance, and processing roles, legal grounds, data subject rights, regulatory requirements, and implementation-oriented traceability. The ontology is evaluated through competency-question-based verification and scenario instantiation. The evaluation provides initial evidence that the ontology makes explicit distinctions that are frequently conflated in legal and technical discourse and provides a reusable conceptual foundation for requirements engineering, regulatory comparison, and compliance-oriented modeling of information systems that support TPDFs.

Alexandra Rowland, Glenda Moura Amaral, Tiago Prince Sales, Giancarlo Guizzardi and Rob Lemmens
Access control, privacy engineering, and confidentiality frameworks are fundamentally concerned with managing who knows what and depend on the ability to reason about the active maintenance of ignorance, yet none provides a formal ontological account of the socialintentional phenomenon that underpins them. Without this, these systems can specify the normative surface of concealment without being able to reason about the beliefs, goals, and commitments that constitute and sustain it. This gap is addressed by taking a social-intentional approach and modelling secrecy not as an epistemic state but as an actively produced and maintained condition grounded in the intentional and social structure of the agents involved. This paper presents a formal ontology of secrecy grounded in the Unified Foundational Ontology (UFO) and specified in OntoUML, supporting interoperability with existing ontologies for trust, value, risk, and prevention. The secret is modelled as a social-intentional relator constituted by a Consequential Belief, or the agent’s goal-grounded assessment of the risk that revelation poses, and a commitment to maintaining nescience, situated within a trust structure derived from ONTrust and a risk and prevention structure grounded in COVER. The resulting ontology distinguishes Secret Owners, Holders, and Keepers; models weak and strong trust as grounding different levels of normative obligation; and characterises secret-keeping as a distributed preventive structure in which every Keeper contributes an antidote to the collective risk of revelation. The result is a reusable ontological module that fills a foundational gap in information systems design by providing the intentional structure that existing frameworks implicitly depend on but do not themselves supply.

Antoine Leblanc, Zequan Huang, Hervé Cholez, Jacques Robin, Sébastien Larinier, Giancarlo Guizzardi, Italo Oliveira, Tiago Prince Sales, Xavier Rousseau, Nourhène Ben Rabah and Bénédicte Le Grand
Cybersecurity is a constantly-evolving domain, driven by the rapid emergence of new vulnerabilities and the continuous expansion of novel technological environments. Amidst this complexity, Security Operations Center (SOC) analysts must continuously assess the risks threatening their infrastructure by reconstructing attack scenarios from scattered signals. Cybersecurity ontologies support this by modeling core concepts and structuring Cyber Threat Intelligence data, yet existing proposals suffer from limited reuse beyond their original use-case and, critically, do not represent the attack plan as an entity that exists independently of the traces it produces. We propose the Reference Ontology for Cybersecurity Attack Planning (ROCAP), which is grounded in the Unified Foundational Ontology (UFO) and aligned with MITRE’s de facto industry standards. By formalizing the attack plan as an intentional structure, ROCAP is designed to be reused and adapted across the evolving landscape while easing interoperability with operational workflows. We demonstrate its practical utility for SOC analysts by answering Competency Questions derived from ISO/IEC 27005 and NIST CSF 2.0.

Fabrício Henrique Rodrigues, Nicolau Santos, Joel Carbonera, Haroldo Rojas de Souza Silva, Mara Abel and João Netto
Representing connectivity in fluid flow systems is a key requirement for ontologies across fields ranging from process engineering to urban infrastructure. The modeling of the dynamic relationship between parts of a complex system that allows data or material to flow across those components remains an open question. While the relation of flow connection is intuitively perceived as transitive and frequently behaves as such, engineering practice reveals numerous scenarios where this property fails to hold, either due to the internal structural isolation of components or their dynamic operational states. This context-sensitive behavior represents a substantial modeling challenge, imposing a trade- off among capturing valid indirect flow pathways, preventing spurious connections, and maintaining model parsimony. This paper proposes the notion of selective transitivity and, based on that, presents a novel Ontology Design Pattern (ODP) to account for this characteristic of the flow connection relation. The pattern is implemented using OWL and SWRL and supports the automated inference of indirect pathways while systematically avoiding unintended, structurally or operationally blocked connections and preserving parsimony. We illustrate the pattern’s application in a case study in the Oil & Gas industry. Given the generality of the notion of selective transitivity, we argue that this ODP can be extended/adapted to other domains.

Remco Dijkman, Arik Senderovich and Stefanie Rinderle-Ma
Business processes often require optimization decisions, in- cluding resource assignment, case scheduling, and path routing. Such decisions are tightly coupled to process structure, resource availability, and uncertain activity durations. Yet the two communities that typically address such problems approach them from opposite directions: business process management (BPM) offers rich process models but limited optimization semantics, while operations research (OR) offers powerful solution methods but often abstracts away process structure early. We address this gap with BPMN+OPT, a BPMN-linked conceptual lan- guage that leaves the standard process structure unchanged and adds an optimization annotation layer. The layer identifies case and resource state used for optimization, stochastic behavior, controllable decision moments and variables, feasibility constraints, and objectives. We give BPMN+OPT an execution semantics based on Semi-Markov Decision Processes (SMDPs), supporting continuous-time, event-driven decision making with non-memoryless durations. This bridges process models and optimization-oriented execution. We demonstrate scheduling, resource assignment, and routing in one process and evaluate a proof-of-concept pipeline that maps BPMN+OPT models to SMDPs.

Nathalie Wolf, Holger Wittges and Stefanie Rinderle-Ma
With growing demand for transparency in the sustainability performance of business processes, understanding how process variants shape sustainability outcomes becomes increasingly important. Process mining provides an empirical foundation for such analyses. However, formal methods for identifying sustainability-aware process variants are still lacking. To address this gap, the present article introduces a structured approach for sustainability-aware process variant discovery, which combines trace clustering with execution-level sustainability analysis. The approach is centered on a formal definition of sustainability-aware process variants and integrates sustainability as an additional analytical dimension into process variant analysis. It further enables statistical characterization of process variants and the derivation of sustainability- annotated process models, facilitating their structured comparison. An empirical evaluation on sustainability-annotated event logs demonstrates the applicability and analytical value of the proposed approach and highlights its potential to support sustainability-aware process analysis.

Qingtan Shen, Artem Polyvyanyy, Nir Lipovetzky and Timotheus Kampik
Agent system mining is a branch of process mining that adopts a bottom-up approach to data-driven analysis of socio-technical systems, such as organizations executing business processes. In this setting, a system is analyzed as a collection of autonomous agents, each with its own local state and behavior, that interact with one another and with the environment to contribute to the global behavior of the system. The state-of-the-art approach to discovering systems of agents constructs modular agent system models captured as Petri nets, but does not directly support the analysis of intra- and inter-agent handovers and their frequencies. This paper presents agent-based Directly-Follows Graphs (DFGs), a novel representation of systems of agents that captures the frequencies of intraand inter-agent handovers and annotates edges with handback and agent availability information. This representation can support the analysis of agent-system behavior and identify opportunities for process improvements. Building on this representation, we propose a technique for discovering agent-based DFGs from event data recorded during the execution of business processes. The technique is evaluated using real-world event data by assessing the quality of discovered models in terms of precision and recall. We compare agent-based DFGs with existing representations and show how they can support agent-related analytics, including identifying potential handover-related bottlenecks and understanding collaboration patterns and task interruptions.

Isadora Valle Sousa, Tiago Prince Sales, Eduardo Guerra, Luiz Olavo Bonino da Silva Santos, Henderik Proper and Giancarlo Guizzardi
Although previous studies have identified several goals associated with modeling activities and model use, recent discussions on AI-supported software development have reinforced the need to better understand how modeling generates value in practice. This study investigates how practitioners perceive the value generated throughout modeling activities. To this end, we conducted an empirical investiga- tion combining an online survey and semi-structured interviews involving practitioners working with domain modeling across multiple project contexts. Based on an inductive thematic analysis, the findings suggest that modeling value also emerges from the socio-cognitive activities supported throughout modeling processes. Five major mechanisms through which modeling generates value were identified, revealing how modeling supports negotiation of meaning, cognitive externalization, sensemaking, and coordination throughout software and domain engineering activities. The study contributes empirical evidence regarding how the value of modeling is perceived in practice and advances a process-oriented perspective on modeling value beyond artifact-centric interpretations.

Franziska Hollauf and Johann Eder
Time is a basic category for modeling a universe of discourse and reasoning about its properties. For processes, event-based temporal concepts allow the representation of admissible sequences of events and temporal constraints between events. Temporal constraint networks in various forms provide means for formally modeling temporal requirements of processes and for reasoning about possible conflicts between these requirements. ST N U , Simple Temporal Networks with Un- certainty, express constraints between the timepoints of events and capture the temporal uncertainty of durations depending on external events. Recently, ST N U O, Simple Temporal Networks with Uncertainty and Oracles, have been introduced to explicitly model the timepoint when these durations are known. Agile controllability, a generalization of the well-established notion of dynamic controllability, leverages this knowledge to ensure the absence of conflicts among temporal requirements. Here we introduce ST N U w , a subclass of ST N U O extending ST N U with the concept of weak contingent activities and timepoints, which is sufficient for modeling a plenitude of scenarios while it still admits checking agile controllability with polynomial complexity. We formally define ST N U w and a procedure for efficiently and effectively checking agile controllability of ST N U w . A proof-of-concept implementation facilitates an evaluation of the scalability of the approach.

Maximilian König, Arsalan Ghasemi, Simon Nowack, Anjo Seidel, Jan Niklas van Detten and Mathias Weske
Object-centric process mining enables comprehensive analysis of information systems by representing different perspectives on sys- tem behavior through object types. Such behavior is captured in object- centric event logs, where events are correlated via co-evolving objects of different types. Currently, these object types are treated as conceptually homogeneous in analyses, even though objects represent quite different entities involved in business processes, such as data objects and resources. Based on a literature survey, this paper argues that data objects and resources exhibit fundamentally different behavioral characteristics. These differences, along with organizational roles that resources can play, significantly hamper the analysis of object-centric event logs. To bring this perspective into object-centric process mining, we extend the metamodel of object-centric event logs with the concepts of resources and roles. Additionally, we introduce methods for identifying resources and roles in existing object-centric event logs, thereby enabling their transformation into the novel, extended metamodel. To evaluate the approach, we introduce a prototypical implementation and apply it to a set of publicly available event logs.

Servando V. Pizarro Martinez, Timon T. Aldenhoff, Arnold F. Arz von Straussenburg and Dennis M. Riehle
International aid organizations use hazard-and-threat ontologies to sort incident reports into shared categories for reporting and security-risk management. In volatile settings, these ontologies fall behind emerging threats such as drone incidents, maritime attacks, new violent actors, or early signs of political instability. We frame this as continuous adjustment: refining a deployed ontology over time based on new evidence, without rebuilding or directly editing it. Following Design Science Research, we design an LLM-assisted pipeline that extracts candidate concepts from incident emails, compares them against a read-only baseline ontology, and groups validated candidates into ranked re- view clusters by recurrence and evidence strength. A review application then lets domain experts accept, reject, or place a candidate on Watch, an intermediate state for recurring but unclear signals, so that human judgment stays between LLM output and any safety-related change. The artifact is evaluated through retrospective simulation over 37,204 incident emails at a European development cooperation organization.

Shilpi Gupta, Monalisha Ojha, Mohammad Sadeghi and Colin Atkinson
OntoUML and Multi-Level Theory (MLT) give conceptual models a checkable ontological semantics. However, the feedback produced when checking a model against these semantics is typically tied to the artifact that drives the check, such as class-level constraints, generated simulations, external counterexamples, or transformed Alloy and OWL representations. As a result, explanations become difficult to follow whenever the relevant evidence spans several classification levels and heterogeneous artifacts. We address this problem with a level-agnostic approach in which model content at every classification level, together with the structures that classify, specialise, and relate it, shares one typed graph, over which a safe, stratified Datalog layer derives both reusable semantic relations and diagnostic facts. Rather than returning only textual error messages, diagnostics are reported as explicit witness tuples whose arguments identify the authored model references, profile terms, and typed values that explain each warning or violation. The approach is illustrated using OntoUML well-formedness rules, the BinOver anti-pattern with an explicitly authored individual witness, non-adjacent classification errors, and MLT-style powertype and partition constraints.

Jasmine Hong, Surabhi Nigam, Ida Someh, Sonja Zillner and Shazia Sadiq
The rapid adoption of enterprise AI has been accompanied by a fragmented and evolving landscape of regulations, standards, and policy frameworks. Although these instruments often overlap in intent, they diverge in terminology, scope, and enforceability across jurisdictions, creating significant governance, risk, and compliance (GRC) challenges for organisations. This paper proposes a conceptual model that can provide ontological support for enterprise AI GRC. Following a Design Science Research approach, we analyse thirteen global AI regulations and standards to systematically examine how GRC concepts are covered. Using thematic analysis and structured abstraction, we derive a coherent set of governance, risk, and compliance dimensions that underpin the model. The resulting artefact aims to enhance semantic cohesion, clarify attribution to source instruments, and support cross-jurisdictional reasoning.

Walid Fdhila
Digital identity wallets are becoming critical infrastructure. The European Digital Identity Wallet (EUDIW), mandated by eIDAS 2.0 and its implementing acts, is expected to serve hundreds of millions of European citizens. Yet neither the core regulation nor its supporting technical frameworks (e.g., ARF) provide a consolidated view of the trustworthiness requirements such systems entail. We present a conceptual model of trustworthiness requirements for decentralized digital identity ecosystems, instantiated for the EUDIW. The meta-model organizes the requirements into seven interdependent trust domains, connected through a typed relation language, all grounded in international standards and specifications. The taxonomy comprises and cross-validates 199 requirements with 81 typed relation instances.

Meenakshi Manjunath and Marian Daun
Robotic applications increasingly integrate AI-based components to expand system autonomy and enable more flexible operation in new environments. However, these capabilities also introduce fundamental challenges for requirements engineering, regarding limited explainability and ever-evolving operational conditions. In particular, AI-based components often behave as black boxes, making it difficult for humans to understand why a system acts in a certain way or why a specific decision was made. This lack of transparency becomes especially critical in human-robot interaction, where humans and robots operate in close proximity and depend on each other during shared collaborative tasks. Therefore, safety concerns resulting from AI use need to be identified and sufficiently mitigated during requirements engineering. To support in this, we propose a goal modeling extension that integrates safety and explainability modeling for collaborative robotic AI systems. Use case based investigations show the applicability of our approach and shed light onto the interplay between robotic AI-components, safety, and explainability, which can improve future safety assessment of robotic systems.

Jeshwitha Jesus Raja and Marian Daun
Modern robotic systems are designed to perform a range of tasks, making their early and precise requirement specification essential. At the same time, current robotic systems heavily rely on standardized robotic hardware, i.e., an application is developed for a specific already existing off-the-shelf robot. While system requirements are typically defined early in the design process, kinematic considerations are often only introduced at later stages, which can lead to infeasible designs and costly redesigns. Therefore, in this paper, we propose the early specification of kinematic robot requirements. Kinematic requirements are derived from high-level system requirements, the technical robot specification containing kinematic properties, and general robot capabilities. This approach allows for early assessment of a robot’s motion capabilities, helping to identify potential limitations before physical development. We evaluate our approach by applying it to a current industry challenge as use case.

Rodrigo F. Calhau, João Paulo A. Almeida, Giancarlo Guizzardi and Victorio Albani Carvalho
Systems are usually conceived of as ‘complex’ (or ‘organized’ wholes) composed of ‘connected’ (or interacting) elements. They are dynamic entities whose composition, structure, aspects, and environment may change over time. In this paper, we take a step towards ontological foundations for modeling system change. Building on the Unified Foundational Ontology (UFO), we account for different dimensions of system variation over time inspired on Bunge’s CESM model. The proposed foundations contribute to system change understanding and provide a conceptual base for future applications in system modeling approaches and languages.

Paolo Bresciani, Paolo Giorgini, John Mylopoulos, Anna Perini and Angelo Susi
Agent-Oriented Software Engineering methodologies such as Tropos were developed under assumptions inherited from symbolic models of agency, including stable goal interpretations, explicit reasoning, and a clear separation between design time and run time. The emergence of Agentic AI challenges these assumptions. LLM-based agents interpret goals at runtime, adapt through inter- action and feedback, and exhibit behaviour that cannot be completely specified during design. In this paper, we argue that these changes require revisiting the conceptual foundations of Tropos rather than replacing its intentional perspective. We introduce a set of conceptual extensions, including the distinction between design-time and run-time intentionality, normative goals, delegation of reasoning authority, residual responsibility, and dynamic autonomy. We then discuss how these concepts may influence the interpretation of the Tropos development process and outline a research agenda for extending goal-oriented requirements engineering towards Agentic AI systems. Rather than presenting a complete methodology, this work aims to stimulate discussion on how intentional modelling should evolve to support adaptive, accountable, and continuously evolving intelligent systems.

Léna Bamouh, Hugo Bruneliere, Jean-Marie Mottu and Massimo Tisi
The engineering of large and complex systems often involves many heterogeneous and sometimes repetitive tasks. Thus, technical solutions are needed to automate such tasks and related workflows. The advent of Large Language Models (LLMs), associated with task-specific agents, offers promising opportunities. However, the orchestration of workflows that involve the use of multiple (LLM-based) agents can be challenging: How can we precisely model such workflows and execute them in practice? In this paper, we propose relying on a specialization of SysML v2 to support the definition of multi-agent workflow orchestrations. We envision a generic orchestration framework that allows for the execution of these SysML v2 models and underlying workflows. In the context of the MOSAICO project, we show that our SysML v2-based orchestration language is expressive enough to model different workflows in various domains. We present initial experiments on such an orchestration framework to execute these workflows by calling LLMs whenever needed.

Luca Franziska Hörner, Maximilian Möller and Manfred Reichert
Large language models (LLMs) have shown substantial potential for generating BPMN 2.0 process models from textual process descriptions. However, systematically assessing whether generated models preserve the underlying process semantics remains challenging due to the limited availability of scalable semantic evaluation approaches. This work proposes a round-trip transformation approach based on bidirectional text-to-model and model-to-text transformations to operationalize semantic fidelity in LLM-generated BPMN 2.0 process models. The approach combines complementary text- and process-model-based analyses to assess semantic completeness and semantic correctness. An empirical study comprising 180 generated process models indicates consistently high semantic completeness, while preserving temporal and logical ordering relations remains considerably more challenging. Furthermore, semantic fidelity proved robust across the evaluated LLM- and prompting- based generation settings, although prompting strategies significantly affected precision-related semantic completeness. Overall, the proposed approach enables systematic and comparatively scalable semantic fidelity evaluations of automated process model generation.

Anastasiya Damaratskaya, Catherine Sai and Stefanie Rinderle-Ma
Regulatory texts often include process-relevant and organizational information, such as decision logic, duties, and actor relations. However, prior research in legal information extraction seldomly addresses organizational structures, and current text-to-process methods are not designed for raw regulatory text. Unlike process descrip- tions, regulatory documents exhibit greater linguistic complexity and non-sequentiality, often applying implicit actors and cross-references that hinder information retrieval. This paper presents a framework that automatically extracts organizational and process-relevant information from regulatory texts and generates machine-readable artifacts, such as organizational models, role-task mappings, and process descriptions. Thus, the approach yields process-oriented representations from legal text while capturing the organizational context necessary to operationalize regulatory requirements. The quality of the generated artifacts and their utility for downstream process model generation tools are evaluated for ten European regulatory documents. The results indicate that the approach outperforms LLM-only baselines, improves the usability of regulatory text, and enables the extraction of organizational and process-relevant information with promising quality for further automation and analysis.

Wim Laurier and Satoshi Horiuchi
Business transactions are fundamental to contemporary information systems, yet their operational implementation often lacks a precise, executable specification of transaction states and lifecycles. Although the Resource–Event– Agent (REA) ontology provides a strong semantic foundation for modeling economic phenomena, it offers limited support for representing dynamic state changes and end-to-end transaction behavior. This paper introduces STREAM (Status of Transaction REA–MERODE), a comprehensive transaction status specification that formalizes business transaction lifecycles as a synchronized set of finite state machines grounded in REA and implemented using the MERODE model-driven engineering methodology. STREAM explicitly defines transaction states and transitions across all phases identified in ISO/IEC 15944—planning, identification, negotiation, actualization, and post-actualization. In contrast to prior REA- or ISO-based approaches, the specification systematically incorporates iterative offer–counteroffer negotiation as well as unsuccessful transactions, rollbacks, and restarts. The resulting model could serve as a reusable and extensible template for developing standardized, interoperable, and ontology-driven business information systems.

Álvaro Bernal-Caunedo, Marco Franceschetti, Hagen Völzer, Manuel Resinas and Barbara Weber
Process mining relies on event data to analyze and improve business processes. The exploratory and non-linear nature of these analyses makes it difficult to understand, validate, compare, and communicate how findings are produced, limiting their business value. Existing provenance techniques focus on data lineage but lack support for systematic comparison of analyses, modular reuse of analysis steps, and comprehensive execution tracking. To address these limitations, we systematically derive a set of requirements and acceptance criteria based on established analytic provenance purposes. We further contribute a structured comparison of state-of-the-art provenance models, revealing critical gaps in comparative capabilities, reusability, and execution tracking. Based on these insights, we propose a conceptual provenance model tailored to exploratory process mining that supports branching analysis paths, comparison of analysis histories and states, reusability of analysis steps, and transparent representation of data evolution.

Marisol Barrientos, Karolin Winter and Stefanie Rinderle-Ma
Ensuring regulatory compliance of business process models remains challenging, with existing approaches focusing primarily on violation detection rather than resolution. To address this gap, this paper presents the Process Tree Repair (PTRepair), a framework for generating, applying, and assessing resolution strategies for compliance violations in business process models. PTRepair combines Large Language Model-based generation of candidate resolution strategies with rule-based process model transformations to ensure consistent and reproducible adaptations of process trees. The generated strategies are then assessed based on structural and behavioral soundness, resolution effectiveness, contextual consistency, change cost, and change location, enabling users to compare and select the most suitable alternative. An evaluation across business process models from different application domains demonstrates that PTRepair generates valid, high-quality resolution strategies that support compliance-aware process redesign.

Clara Kowalczyk, Alexis Brissard, Frederic Cuppens and Amal Zouaq
Evaluating the quality of business process models generated by Large Language Models (LLMs) is critical, yet existing approaches rely on ad hoc metrics or comparison to a gold-standard model. We argue this assumption is restrictive, as multiple process models can accurately represent the same process. We propose PETALE, a fine-grained evaluation framework that extract both process models and natural-language description into fine-grained semantic units (tasks, actors, etc.), and aligns them via semantic similarity to enable precise, reference model-free identification of errors and omissions. To support framework valida- tion, we introduce TaskPET, a dataset of 45 manually-annotated processes using this units representation. Benchmark of state-of-the-art automated process modeling approaches against human expert baseline on TaskPET reveals modeling deficiencies of existing holistic metrics: while most methods reach near-human performance at task-level, they tend to fall short in identifying finer-grained elements such as actors or resources.

Benedikt Sennebogen, Antoon Bronselaer, Toon Boeckling, Uta Störl and Meike Klettke
In many areas of computer science (such as modeling and schema languages), both simple, easy-to-learn methods and complex, powerful solutions are simultaneously available. The simple solutions are often used when interacting with users. The currently available JSON schema languages are highly expressive, but not easy to write manually. Therefore, we propose an alternative: a very intuitive yet expressive schema language for JSON, called JSON Document Type (JDT) Schema Language. In addition to features for defining structural constraints, simplicity, similarities to the English language, and compactness have been key design criteria. In this demo, we showcase the JDT language and two different tools: an editor that can be used to easily define a JDT schema and a tool to validate JSON documents against JDT. The target audience are researchers with different backgrounds looking for an intuitive method to define the structure of their JSON documents to streamline their data processing applications.

Andrea Colombo
Historical banking archives contain rich evidence for studying financial institutions, governance, and political economy, yet their analysis remains severely constrained by heterogeneous document formats, degraded scans, and evolving institutional structures. We present an end-to-end conceptual modeling and data engineering framework that transforms 182,719 scanned pages of Italian banking supervision records from 1910–1960 into a structured, document-centric relational database. The framework combines Vision Language Models and Large Language Models in a six-stage batch-oriented ETL pipeline covering OCR, document segmentation, metadata classification, data integration, governance extraction, and credit exposure extraction. Its conceptual model comprises a central document entity linked through explicit foreign keys to governance entities and credit exposures, balancing expressive representation of historical relationships with the practical constraints of automated LLM-based extraction. The resulting database contains more than 430,000 records spanning 778 banking institutions, including 34,771 governance entities and 212,760 credit exposures.

Naomi Feldmann, Daniel Borcard, Hans-Georg Fill and Simon Curty
Conceptual models are widely used to design and control robotic tasks. Existing modeling environments in robotics are usually two-dimensional and separated from robot avatar simulations. This makes it difficult to represent spatial aspects of robotic behavior such as the robot’s position and movement. We introduce a browser- based virtual reality environment that combines immersive Petri net modeling, execution, and synchronized robot behavior visualization. Users can create, modify, and execute Petri net models directly in virtual reality. The execution of the Petri net is connected to a robot avatar, which performs corresponding actions such as movement, pick, and place operations in real time. This way users can directly observe how the execution of a conceptual model translates into robot behavior.

Luca Franziska Hörner, Maximilian Möller and Manfred Reichert
Conceptual modeling is increasingly supported by Large Language Models (LLMs) through conversational interaction. In the domain of business process modeling, existing approaches have demonstrated the feasibility of generating BPMN 2.0 process models from natural language. However, their support for collaborative process elicitation, multi-perspective BPMN 2.0 modeling, and process understanding remains limited. This paper presents BPMNGen, an LLM-based conversational environment that enables brainstorming-based BPMN 2.0 modeling, as well as the generation of collaboration, choreography, and conversation BPMN 2.0 diagrams. Moreover, BPMNGen supports transformations between these BPMN perspectives and BPMN-to-text transformations to foster process understanding and documentation. BPMNGen provides hands-on insights into LLM-assisted collaborative business process modeling and will be of particular interest to researchers and practitioners working on conceptual modeling, Business Process Management (BPM), and generative AI.

Johannes Löbbecke, Marisol Barrientos, Karolin Winter, Juergen Mangler and Stefanie Rinderle-Ma
Adhering to compliance constraints is essential for all companies regardless of size or domain. Business Process Compliance (BPC) uses Business Process Management (BPM) techniques to ensure that a company’s processes adhere to compliance constraints. However, existing approaches generally address only one or a few aspects of BPC, such as requirements extraction or model verification, and typically have low technology readiness. We present a compliance suite integrated into a process execution engine that supports extracting constraints from natural language and using them to verify and repair the compliance of process models. The resulting suite combines the automation capabilities of LLM-based requirements extraction and compliance repairing, with the trustworthiness of deterministic verification and engine-integrated modeling, simulating, and execution.

Luís Pedro Baptista, Vânia Sousa, Pedro Guimarães, António Amaro Costa Vieira and Maribel Yasmina Santos
Industrial digital transformation generates a lot of data, but turning it into useful insights for business strategy remains difficult. Traditional data-driven methods often overlook what decision-makers need, while requirements- driven methods can suggest solutions that are impractical due to insufficient data. To align these two perspectives, this paper presents a web-based prototype that uses a multi-driven analytical modeling approach supported by Large Language Models (LLMs). Our tool automatically proposes Data Warehouse (DW) schemas and dashboards by connecting physical data models to requirements structures from the iStar framework. Data engineers can review and improve the results generated by the LLM through a Human-in-the-loop process. Used in real-world case studies from industry, this combined Human-AI approach accelerated DW design, improving the connection between strategic goals and analytics, and also helped identify key data gaps.

Jeshwitha Jesus Raja, Shaza Elbishbishy, Yanire Gutierrez, Philipp Kranz and Marian Daun
Collaborative robots are increasingly adopted in manufacturing to improve flexibility and productivity, but seamless human-robot collaboration requires mechanisms that prevent errors and ensure safety. Unlike conventional 3D-based digital twins that simulate the physical system, a model-based digital twin captures the logical flow of tasks, dependencies, and interactions, offering a lightweight and scalable solution for adaptive monitoring and control. This paper presents a model-based digital twin that reuses design-time artifacts, namely Goal Models and Process Models, for runtime monitoring, task synchronization, and error detection. The framework is evaluated on a collaborative assembly process.

Meike Ullrich, Fabian Aisenbrey and Andreas Fritsch
Providing timely and meaningful feedback on student-created conceptual models requires assessing syntactic and semantic correctness, the latter indicating whether a model adequately represents the intended domain. Existing rule- and label-based approaches provide limited support for alternative valid terminology and modeling decisions, while direct large language model (LLM) assessments may produce incomplete or inconsistent feedback. This paper presents an LLM-assisted semantic assessment approach for Entity–Relationship diagrams that combines explicit diagram mapping and deterministic difference detection with LLM-based semantic interpretation. The approach generates element-specific feedback and makes identified differences between the task description, the sample solution, and the submitted model transparent. It supports configurable assessment instructions and generates detailed formative feedback. Evaluation on student submissions shows that the assessment approach with explicit diagram mapping and deterministic difference detection produces substantially more complete and correct feedback than direct LLM assessment.

Andrea Avignone, Silvia Chiusano, Alessandro Fiori, Sara Giorza, Santa Panduri and Riccardo Torlone
Database design is a crucial practice in relational databases, following well-established patterns. The rising popularity of document-oriented NoSQL databases has changed how data modeling is perceived due to their flexible schema. As a consequence, data modeling is often reduced to an application-driven approach, where conceptual design is not addressed and schema decisions are tightly coupled with implementation details. Recent work has shown that Entity-Relationship (ER) modeling principles can be adapted to design document schemas, but practical tool support remains limited. We present NoER, a web-based system that supports a conceptually grounded and guided design process for document-oriented databases using a revised version of the ER model. It guides users through an end-to-end workflow, from conceptual schema definition to the interactive exploration of translation alternatives and the automatic generation of a MongoDB-compatible JSON schema. The demonstration showcases how NoER helps users compare design options, apply document design patterns consistently, and derive document schemas that are aligned with application requirements and easier to maintain.

Shivam Sharma, Mengxiang Jiang and Shraddha Sharma
We present a conceptual model for a knowledge management system (called LIST ), its reference implementation as a Unix-based filesystem (called LFW ), and utilization of this model for applications beyond file management, e.g. a git middleware (called git_lfw) which preserves file provenance automagically. In this paper, we describe how this conceptual model was born from a software project for unifying data sources (phase-1) to the current filesystem implementation (phase-2) and finally to a conceptual model specification (phase-3). The work advances conceptual modeling by providing a practical multi-model abstraction for knowledge representation, and in-line with ER-2026 theme on “Sustainability”, this work enables sustainability in two key ways: (1) sustainability of knowledge management for a human lifespan, and (2) environmental sustainability. More details about the project can be found at: https://listspec.org.

Joshua Send
Fifty years on, Chen’s Entity-Relationship model is still translated into poorer implementation models—losing relationship identity, subtypes, roles, and participation constraints along the way. This tutorial executes a conceptual model directly in TypeDB, whose language TypeQL implements a higher-level polymorphic entity-relation-attribute model grounded in type theory (Best Newcomer Award, ACM SIG- MOD/PODS 2024). Building a sustainable supply-chain model, participants progress from a classic ER baseline to polymorphic querying, n-ary supplier networks, and recursive traceability—and leave able to turn an ER diagram into a running, schema-validated database.

Bert de Brock

Sander Leemans

Zixin Wu

Samaneh Kiamanesh

Saba Malekpour Bejandi