The evaluation of modeling languages for augmented reality applications poses particular challenges due to the three-dimensional environment they target. The previously introduced Augmented Reality Workflow Modeling Language (ARWFML) enables the model-based creation of augmented reality scenarios without programming knowledge. Building upon the first design cycle of the language's specification, this paper presents two further design iterations for refining the language based on multi-faceted evaluations. These include a comparative evaluation of implementation options and workflow capabilities, the introduction of a 3D notation, and the development of a new 3D modeling environment. On this basis, a comprehensibility study of the language was conducted. Thereby, we show how modeling languages for augmented reality can be evolved towards a maturity level suitable for empirical evaluations.
Domain modeling is typically an iterative process where the modeling expert interacts with the domain experts to complete and refine the model. Recently, we have seen several attempts trying to assist, or even replace, the modeler with a Large Language Model (LLM). Several prompting strategies have been attempted but with limited success so far.
In this paper, we advocate for the adoption of a Tree-of-Thoughts (ToT) strategy to overcome the limitations of current approaches based on simpler prompting strategies. With ToT, we can decompose the modeling process into several substeps using for each step a specialized set of generators and evaluators prompts to optimize the quality of the LLM output. As part of our adaptation, we provide a Domain-Specific Language to facilitate the formalization of the ToT process for domain modeling. Our approach is implemented as part of an open source tool availabe on GitHub.
On top of database technologies on diverse data models, the concept of polystore has been studied for unified management of heterogeneous data. However, the design of multi-model query languages, which are the major interface of polystores, is an important yet challenging task. There are still drawbacks in the existing multi-model query languages, including limited expressiveness of query representation and fixed data models.
To address these problems, this paper presents the Equivalently Combinations of Query Languages (ECQL) as a holistic representation of multi-model queries. This paper firstly makes fundamental abstractions of the query results of different data models, and then designs a new multi-model query language that combines the query languages of different data models based on these abstractions. We explain the expressiveness of ECQL that it supports both single-model query functionalities and cross-model queries. The evaluation results between ECQL and existing multi-model query languages on polystore benchmarks show the succinctness of ECQL. We also demonstrate the extensibility of ECQL that it can be integrated with new type of query languages.
In the last decades, Model-Based Systems Engineering (MBSE) has received significant attention, leading to standards such as SysML. SysML is due to a recent and radical update, breaking the dependence of its specification from UML, and leading to the development of two languages: KerML, which provides a top layer of general constructs, and SysML v2, which specializes KerML for systems engineering. In this paper, we analyze the formal and real-world semantics of the proposed KerML and SysML v2 specifications, and draw implications for their improvement and further development. Our attention is focused towards key constructs of the languages. We also examine the approach taken in the specifications to deal with dynamic aspects, which is inspired by the 'four-dimensionalist' view.
In this work, we propose a conceptual model for existential events that affect portions of matter. Portions of matter are entities that constitute most of the physical entities, for instance, water, oil, gold, and blood. Existential events create and destroy entities. The proposed conceptual model is intended to support the design of information systems that capture the dynamic changes of portions of matter in nature and industrial processes. First, we propose a taxonomy for Collections of Objects (the entities that constitute portions of matter) and define the phase transitions that create and terminate the circumstances for a portion of matter to exist. Second, we define three existential events that affect portions of matter: (i) Forming Stuff, in which an instance of portion of matter is created as a collection of objects become connected; (ii) Terminating Stuff, in which an instance of portion of matter is terminated as a collection of granules become scattered; (iii) Changing Stuff Type, where a quantity of one kind is created and a quantity of another kind is terminated.
Lastly, we present an application example that uses the condensation of a portion of water and its freezing into a portion of ice to illustrate how to use the conceptual model in concrete cases.
Reinforcement learning (RL) is an important class of machine learning techniques, in which intelligent agents optimize their behavior by observing and evaluating the outcomes of their repeated interactions with their environment. A key to successfully engineering such agents is providing them the opportunity to engage in a large number of such interactions safely and at a low cost. This is often achieved through developing simulators of such interactions, in which the agents can be trained while also different training strategies and parameters can be explored. However, specifying and implementing such simulators can be a complex endeavor requiring a systematic process for capturing and analyzing both the goals and actions of the agents and the characteristics of the target environment. We propose a framework for model-driven goal-oriented development of RL simulation environments. The framework utilizes a set of extensions to a standard goal modeling notation that allows concise modeling of a large number of ways by which an intelligent agent can interact with its environment. Though subsequent formalization, the model can be used by a specially constructed simulation engine to simulate agent behavior, such that off-the-shelf RL algorithms can use it as a training environment. We present the extension of the goal modeling language, sketch its semantics, and show how models built with it can become executable.
Identifying and analyzing security requirements is an essential part of the information systems engineering lifecycle. Several techniques have been introduced for comprehensively modeling such requirements. Once identified, security requirements must be translated into designs that allow domain actors to securely accomplish business tasks under given risk assumptions and contexts. Correctly translating requirements to such designs, however, can be challenging when considering both the complexity and specialized nature of security mechanisms, such as cryptography, and the role of varying practical and contextual aspects of the problem at hand in correctly applying such mechanisms. We propose a model-driven pattern-based approach for supporting the implementation of security requirements. Security requirements models, augmented with descriptions of contextual and threat assumptions, are combined with reusable domain-agnostic workflow patterns which model established ways for securely performing common business tasks. The combined models are compiled into a formal specification, whereby automated reasoning is applied for generating domain-appropriate workflows that satisfy the security requirements. Using the technique, analysts can efficiently explore the impact of different threat assumptions and pragmatic constraints to candidate security designs, while ensuring that the latter are consistent with tried-and-tested community knowledge.
Domain modeling involves creating abstract representations of information within a specific domain using techniques such as conceptual modeling and ontology engineering. Traditionally, manual creation and maintenance of domain models are labor intensive and require modeling expertise. This paper explores the automation of domain modeling using pre-trained large language models (LLMs), presenting an experimental LLM-based conceptual modeling assistant that collaborates with a human expert. The assistant provides modeling suggestions based on a given textual description of the domain of interest, aiding in the design of classes, attributes, and associations. We present a generic framework for domain modeling assistants that consists of class, attribute, and association generators, and show how they can be implemented using an LLM. We demonstrate a concrete configuration of this framework and its implementation in a prototype application. We evaluated the effectiveness of the framework configuration across various domains. Our findings indicate that the assistant significantly enhances the efficiency of modeling while maintaining reasonable quality of the outputs.
Large Language Models (LLMs) have opened new opportunities in modeling in general, and conceptual modeling in particular. With their advanced reasoning capabilities, accessible through natural language interfaces, LLMs enable humans to deepen their understanding of different application domains and enhance their modeling skills. However, the open-ended nature of these interfaces results in diverse interaction behaviors, which may also affect the perceived usefulness of LLM-assisted conceptual modeling. Existing works focus on various quality metrics of LLM outcomes, yet limited attention is given to how users interact with LLMs for such modeling tasks. To address this gap, we present the design and findings of an empirical study conducted with information systems students. After labeling the interactions according to their intentions (e.g., Create Model, Discuss, or Present), and representing them as an event log, we applied process mining techniques to discover process models. These models vividly capture the interaction behaviors and reveal recurrent patterns. We explored the differences in interacting with two LLMs (GPT 4.0 and Code Llama) for two modeling tasks (use case and domain modeling) across three application domains. Additionally, we analyzed user perceptions regarding the usefulness and ease of use of LLM-assisted conceptual modeling.
This paper investigates how the model size affects the ability of a Generative AI Language Model, or briefly a GLM, to support the text-to-SQL task for databases with large schemas typical of real-world applications. The paper first introduces a text-to-SQL framework that combines Prompt engineering and a Retrieval-Augmented generation (RAG) technique, leaving as flexibilization points the GLM and the database. Then, it describes a benchmark based on an open-source database with a large schema, with a complexity similar to real-world databases. The paper proceeds with experiments to assess the performance of the text-to-SQL framework instantiated with the benchmark database and GLMs of different sizes. The paper concludes with recommendations to help select which GLM size is appropriate for a text-to-SQL scenario, characterized by the difficulty of the expected NL questions and the data privacy requirements, among other characteristics.
Software Engineering research aims to enable the effective transformation of stakeholders' requirements involved in the software development life cycle into code such that purpose of the code is to fulfil the requirements. Crucial software maintenance and evolution tasks such as change impact and software reusability require an explicit linkage between the requirements and their code manifestation. However, manually creating such links for large codebases is a prohibitive task. Furthermore, the requirements are given in natural language (NL) and therefore contain natural language semantics that capture the purpose of the corresponding codebase. Existing traceability approaches rely on utilizing textual similarities between requirements and code to establish links, however, such approaches tend to achieve low precision due to their inability to bridge the semantic gap between high-level natural language requirements and syntactic code. With the advent of Large Language Models (LLMs), their sophisticated capabilities to comprehend both natural language and code syntax offer a transformative potential for bridging this semantic gap. LLMs' ability to understand the underlying intent of NL requirements and their corresponding code snippets can significantly enhance the precision of traceability links. By representing code as a knowledge graph, this framework can also leverage graph structural information to further refine and strengthen traceability. Therefore, in this paper, we present an LLM-supported retrieval augmented generation based framework for traceability that supports vector, graph, and hybrid (combined) indexing. We provide a comparative evaluation with an existing approach and between different indexing techniques. Finally, we discuss the results of our evaluation, showcasing how our method improves the efficiency and accuracy of establishing traceability links in software development.
The rapid proliferation of Application Programming Interface (API) enhances data exchange but introduces significant privacy and security risks. This issue is particularly acute in the Internet of Things (IoT), where APIs interact with numerous devices, often lacking adequate mechanisms for managing privacy and security on the API and client’s side, leading to increased vulnerabilities. Hippocratic databases provide mechanisms, e.g., role-based access, to control access to the databases. However, to effectively manage data access to the Hippocratic database, proper design of API is needed. This paper proposes a conceptual framework for Hippocratic API (HAPI) inspired by the Hippocratic Oath and GDPR. Our framework revises traditional API designs using AI to integrate ethical considerations, aiming to protect data subjects’ rights and enhance security. By embedding data protection and ethical standards into API operations, HAPIs rectify inadequacies in consent mechanisms and mitigate privacy risks. We leverage an ontology-based approach to identify and categorize non-functional requirements (NFRs) and employ design techniques found through extensive research of literature, books, and tutorials. This methodology, informed by GDPR, ISO/IEC 27001, and Hippocratic databases, employs the Unified Foundational Ontology (UFO) framework to model goals, intentions, and qualities, ensuring ethical principles are intrinsic to the system’s design. We hope HAPIs will inspire further research and encourage both industry and academic communities to refine and expand upon our design, fostering a future where technological development is inherently aligned with the ethical management of user data.
Air Quality Index (AQI) is a number aggregated from several air quality sensors deployed in an area. AQI is useful in communicating the air quality to the general public and in making governance decisions to tackle air pollution. However, our ethnographic surveys revealed the existence of a knowledge barrier in interpreting the AQI and data illiteracy in understanding AQI-related charts and trends commonly facilitated by government organizations. This knowledge gap is wider for the marginalized sections of society, who, it turns out, are more exposed to pollution. We use an ontological approach to homogenize the air quality data with social and spatial aspects. The Social Air Quality Index (SAQI) ontology integrates the data from local and central air quality monitoring sensors, meteorological data, and field surveys. This data is converted into a Knowledge Graph, which is used to build an application for civic engagement with the public on pollution to improve community participation of the local institutions and individuals. We evaluated this application through a user survey and received positive feedback. The ontologies, code, and datasets are available under the Apache 2.0 License at https://github.com/kracr/aq-structured-platform.
Time is a fundamental concept in data processing. The growth of social media (SM) data and the rise of the Internet of Things (IoT) underscores the necessity for studying temporal data on the Web. However, accessing realistic temporal data poses significant challenges across data collection, knowledge representation, and real-time provisioning, with no comprehensive solution available yet. To tackle these challenges, we introduce GenACT, a novel data generator rooted in the dynamics of Academic Conference Tweets (ACT), which serves as an ideal domain for eliciting application scenarios spanning temporality, dynamicity, and timeliness. The foundation of GenACT is a domain-specific ontology crafted to conceptualize tweets around an Academic Conference Event (ACE) realistically. The ACE ontology is available in all four OWL 2 profiles. Additionally, RDF instantiation allows for real-time simulation of ongoing academic discussions on Twitter. GenACT stands out for its ability to configure different data segments using SPARQL-based partitioning strategies. This versatility makes it adaptable to various analytical tasks, enabling researchers to focus on specific aspects of the data for their studies. GenACT is designed to seamlessly provide temporal and static data in a streaming format, tailored specifically for applications in studying knowledge graph evolution, temporal reasoning, and stream reasoning. The ontology, code, and documentation are available under the Apache 2.0 License at \url{https://anonymous.4open.science/r/genACT/}.
The BERT model has slow inference speed due to the massive parameters, and it is susceptible to adversarial attacks due to lack of robustness. Knowledge distillation tackles the first issue by compressing the model, while adversarial training resolves the second issue by incorporating disturbances into training. However, existing knowledge distillation methods often struggle to balance between model accuracy and robustness. To address this, we propose a multi-teacher adversarial distillation approach for the BERT model. Our approach employs a clean teacher, providing soft labels to guide the student model in learning the clean samples, and a robust teacher, which employs adversarial training and semi-supervised learning to guide the student model in learning adversarial samples. Additionally, we introduce adaptive loss weights, which enable the student model to focus on the more challenging knowledge during training process. Through experiments with various adversarial attack methods, we validate the effectiveness of our proposed approach. On the typical text classification dataset IMDB, we improve the robust accuracy of the student DistilBERT model from 19.78 to 35.17, while maintaining a high clean accuracy of 93.17.
Modularization has been extensively investigated for its role in enhancing the comprehension of process models by dividing them into smaller, self-contained and manageable modules. However, existing studies have reported inconclusive results, prompting further exploration into when modularization supports or impedes process comprehension. Among the key factors suggested to influence the effect of modularization is the type of the task at hand. Indeed, the fragmentation of information across several modules caused by modularization can challenge readers' comprehension of process models, especially if the task is not confined to a single module. While the effect of fragmentation has been explored for flow-based tasks, requiring understanding procedural aspects of process models, our work extends to circumstantial tasks, which instead demand comprehending model rules and constraints that rely on the broader context associated with specific process parts. Using eye-tracking, we investigate how the fragmentation of information caused by modularization impacts cognitive integration (i.e., the collection and synthesis of dispersed information) and its subsequent effect on cognitive load and model comprehension. Our findings demonstrate that fragmentation significantly affects cognitive integration in flow-based tasks only. Moreover, our results show that cognitive integration is linked to both cognitive load and readers' comprehension of process models. The outcome of this work provides a detailed model, explaining the task-specific effects of fragmentation. Practically, our findings highlight the need for modeling tools that dynamically adjust their user interface} to reduce cognitive integration demands, thereby enhancing model comprehension.
Over the past decade, extensive research efforts have been dedicated to the extraction of information from textual process descriptions. Despite the remarkable progress witnessed in natural language processing (NLP), information extraction within the Business Process Management domain remains predominantly reliant on rule-based systems and machine learning methodologies. A key reason for this is the data scarcity in the domain, which has so far prevented the successful application of deep learning techniques. However, rapid advancements in generative large language models (LLMs) allow solving many NLP tasks with very high quality, with only a few examples (few-shot), or even just a description of the problem (zero-shot).Therefore, we systematically investigate the potential of LLMs to solve the task of information extraction from textual process descriptions, targeting the detection of process elements, such as activities and actors, and relations between them. Based on a novel prompting strategy, we show that LLMs are able to outperformstate-of-the-art machine learning approaches with absolute performance improvements of up to 8\% $F_1$ score across three different datasets.We evaluate our novel prompting strategy on eight different LLMs, showingit is universally applicable, while also analyzing the impact of certain promptparts on extraction quality. The number of example texts, the specificity of definitions, and the rigour of format instructions are identified as key for improving the accuracy of extracted information. Our code, prompts, and data is available at https://anonymous.4open.science/r/llm-process-generation-2140/README.md and our model generation proof-of-concept at https://anonymous.4open.science/r/pet-to-bpmn-poc-B465/README.md.
Event data is a collection of recorded events that capture performed actions and observed states of business processes supported by information systems. It describes the times of event occurrences, event types, event attributes, and process cases of events identified by one or more objects the events relate to. Process mining uses event data to analyze and improve the processes in organizations. These processes are often performed by actors or agents, such as employees, resources, and systems, in different roles within organizations. In this paper, we present Agent System Event Data (ASED), a new type of event data that describes business processes as interactions of agents. ASED provides a new scope for analyzing individual agents involved in multiple processes, interactions of agents, and systems of agents that enact the processes. We formalize ASED as a conceptual data model, discuss its dimensional data modeling aspects, and argue that event data, in general, benefits from dimensional representation. We review existing event data types and discuss the complementary nature of existing models and ASED. Finally, we validate ASED by demonstrating its ability to express existing business process compliance rules, significantly expanding the scope of compliance analysis addressed by existing data models.
Identity and Access Management (IAM) is a crucial aspect of online interactions, and researchers and practitioners have been exploring this area since the early days of the Internet. Over the years, many digital identity protocols and standards were created. While OAuth 2.0 and OpenID Connect (OIDC) are currently the most widely used and mature protocols, the emerging concept of Self-Sovereign Identity (SSI) and its underlying protocols promise greater user privacy and control. Although much emphasis has been placed on new protocols and implementations, the ontological and formal understanding of digital identity models still needs to be addressed. In this paper, we make three contributions: (i) introduce the Role-Artifact-Function (RAF) framework, which comprises a meta-metamodel and a metamodel; (ii) describe and compare digital identity models through the instantiation of the RAF metamodel; and (iii) discuss and arrive at intriguing conclusions, such as the conceptual equivalence between the well-established family of x509-based protocols and the nascent technologies behind SSI.
Amid rising global challenges, resilience is attracting increased interest across various disciplines. However, significant ambiguities, vagueness, and variations in its definitions present notable obstacles to interdisciplinary communication and practical application. These semantic issues negatively impact not only the applications that rely on these definitions but also undermine the integrity of the conceptual models built from them, rendering them unable to ensure interoperability and clarity for their intended users. Recognizing the need for a robust and clear definition of resilience, this work addresses the complexity inherent in the concept by performing an ontological analysis using the Unified Foundational Ontology (UFO) and creating a sound core ontology model with OntoUML, UFO’s related conceptual modeling language. This research unfolds the conceptualization of resilience by investigating its fundamental categories and related concepts. We explore essential aspects of resilience, examining whether it preexists disturbances or is developed in response to them. We also identify its relational dependencies, establish how it is actualized, and determine the circumstances under which it becomes perceptible. Following our analysis of the ontological nature of resilience, the paper establishes an ontologically well-founded definition of this concept. To illustrate the practical application of our theoretical findings, we present a specific case in the field of production management.
Effective conceptual modelling is critical for accurately representing the physical world in digital twin implementations, yet existing methods often lack comprehensive guidance or overlook key aspects like communication modelling. This research proposes the Entity-Relationship Digital Twin (ERDT) modelling method tailored for industrial digital twins. Building on established Entity-Relationship (ER) models, ERDT extends them with constructs for handling historical data, and defining security interfaces and data flows between physical and virtual components. The method provides a systematic process covering digital twin objective setting, physical component selection, conceptual entity definition, interface specification, and communication modelling. It emphasizes principles like encapsulation and separation of concerns to enhance maintainability. Evaluated against ISO 25010 quality criteria, ERDT demonstrates strengths in functional appropriateness, usability, portability, and maintainability while identifying opportunities for further improvement. A logistics case study illustrates applying the method to a real-world industrial scenario. By bridging conceptual modelling with software engineering best practices, ERDT facilitates developing robust, maintainable digital twins.
