KEOD 2026 Abstracts


Full Papers
Paper Nr: 21
Title:

Deciphering Causal Reasoning in Temporal Knowledge Graphs: A Survey

Authors:

Aadya R. Bharadwaj, Adarsh Arun Hoskere, Aditya Mohan Shenoy, Adarsh Nambiath and Bhaskarjyoti Das

Abstract: Temporal Knowledge Graphs (TKGs) provide a powerful framework for modeling time-evolving events and their interdependencies. However, most existing TKG reasoning approaches remain largely correlation-driven, limiting their ability to capture genuine cause–effect relationships, especially in noisy real-world event streams. This paper presents a structured review of causal reasoning in Temporal Knowledge Graphs, with emphasis on temporal modeling, explainability, uncertainty, datasets, and evaluation practices. We categorize existing approaches into rule-based, neural, uncertainty-aware, and causal discovery paradigms, and critically analyze their assumptions, strengths, and limitations. We further examine real-world event sources such as GDELT and highlight a central benchmark–objective mismatch: common TKG benchmarks and ranking-based metrics measure temporal association and prediction quality, whereas causal reasoning demands verifiable causal validity, evidence grounding, and robustness to temporal uncertainty. Based on this analysis, we identify major research gaps, including the absence of standardized causal evaluation protocols, the difficulty of handling noisy timestamps and reporting bias, and the limited integration of explainability with formal causal reasoning. Finally, we outline future research directions toward reliable, interpretable, and causally grounded Temporal Knowledge Graph systems.
Download

Paper Nr: 41
Title:

A CIM-Based Data Model for Energy Flexibility in Multi-Agent Energy Management Systems

Authors:

Sohvi Virkkula, Rita Beigaitė and Ilkka Niskanen

Abstract: In multi-agent energy management systems, energy flexibility relies on coordinated interactions among heterogeneous distributed energy resources and system actors that adjust energy consumption and/or generation in response to varying external conditions. These interactions involve the exchange of structured information between agents, such as forecasts of flexibility potential, flexibility offers, and control instructions. However, differences in data models, standards, and system implementations create semantic interoperability challenges in the representation and exchange of these messages across agents. This paper presents a data model for energy flexibility in multi-agent energy management systems based on the Common Information Model (CIM). The model is derived using a scenario-based approach in which agent interactions and message exchanges form the basis for defining data requirements, including concepts, classes, attributes, and message structures. These requirements are mapped to CIM and used to define a set of profiles that capture energy flexibility-related interactions. The model is evaluated and iteratively extended using real-world message exchanges. The results demonstrate its relevance in realistic scenarios and provide a foundation for the incremental development and validation of interoperable, standards-aligned data representations for evolving multi-agent energy flexibility systems.
Download

Paper Nr: 81
Title:

PoRTA: A Benchmark for Semantic Alignment between RDF Triples and Portuguese Natural Language

Authors:

Victor Jesus Sotelo Chico, André Gomes Regino, Fernando Rezende Zagatti and Julio Cesar dos Reis

Abstract: Natural Language Generation (NLG) from structured semantic representations plays an important role in Knowledge Graph (KG) verbalization, semantic interoperability, and ontology-aware intelligent systems. However, current development and evaluation of RDF-to-text and text-to-RDF approaches primarily focus on English, hindering progress in underrepresented languages such as Portuguese. This article presents PoRTA (Portuguese RDF-Text Alignment), a benchmark dataset that facilitates semantic alignment between RDF triples and natural language verbalizations in Brazilian Portuguese. We constructed PoRTA from DBpedia triples using an ontology-aware pipeline that includes RDF extraction, semantic filtering, LLM-based verbalization, human curation, and linguistic augmentation. The benchmark consists of 5,751 semantically validated RDF triples across nine knowledge domains, each linked to six natural-language variations generated via controlled paraphrasing. We assess PoRTA in RDF-to-text generation scenarios using both fine-tuned and few-shot large language models. Our experimental results show that the benchmark enables semantically faithful verbalization and strong generalization, while also enhancing linguistic diversity compared to translated benchmarks. Additionally, our findings suggest that PoRTA can aid in semantic consistency analysis, ontology-aware generation, and KG verbalization tasks in Portuguese, thereby advancing multilingual knowledge engineering and Semantic Web applications.
Download

Paper Nr: 82
Title:

ComplianceGPT: A Citation-Contract Answerer for Mechanically Verifiable Compliance Question Answering

Authors:

Wenjia Wang, Seyed Masoud Sadjadi, Naphtali Rishe and Arpan Mahara

Abstract: A compliance question answering system can retrieve the governing clauses and still produce an unreliable final answer. A free-form generator may omit revision-specific qualifications, make claims unsupported by its citations, or supply Organization-Defined Parameter (ODP) values that the standard leaves to the organization. ComplianceGPT addresses this problem for National Institute of Standards and Technology Special Publication 800-53 (NIST SP 800-53) by decoupling the model’s evidence-selection role from deterministic citation-contract construction. Given a bounded, revision-specific evidence window, the model selects source identifiers rather than writing final answer prose; deterministic components then resolve those identifiers against canonical clause records, apply ODP policy, assemble a citation contract, and validate its source, span, and status consistency. In a matched comparison, both methods receive the same question, ordered evidence window, and loaded Qwen2.5 model instance. ComplianceGPT attains offline strict-pass rates of 65.0% versus 25.0% for the generative baseline on Revision 5 (Rev. 5) and 80.6% versus 22.2% on Revision 4 (Rev. 4). On no-profile ODP subsets, it always returns a parameter-required status; exact unresolved-ODP set agreement is 87.3% and 63.2% on Rev. 5 and 4, respectively. The released citation contracts are mechanically verifiable against source-resolution, span-containment, and status-consistency rules; offline scoring further shows higher agreement with the implemented source, coverage, and status rules than the model-written baseline.
Download

Paper Nr: 97
Title:

A Knowledge-Based Decision Support System for Hotel Linen Planning from RFID Movement Events

Authors:

Sandro Canada Rodrigues, João Fernando Nunes Alves, David Aveiro and Fábio Mendonça

Abstract: Hotel linen operations generate item-level Radio-Frequency Identification (RFID) movement logs, but using them for planning requires domain interpretation before forecasting. This work presents a knowledge-based decision support system using 45 months of RFID data from a hospitality property in Madeira, Portugal. Its primary scientific contribution is a reproducible method for converting heterogeneous movement events into semantically valid forecasting targets and explicit planning indicators. A conceptual domain model separates normal laundry returns from discrepancy-correction movements. Monthly product-level targets are then forecast by a Naive lag-1 and Random Forest (RF) ensemble. Across 21 one-month-ahead expanding-window folds, the production model obtains 6.10% mean absolute percentage error versus 6.28% for Naive, a small and uncertain gain. Drop-column ablation shows that RF performance does not depend solely on lag-1, while lagged adjustments add no benefit. On the same corrected target, Seasonal-Trend decomposition using Loess (STL) gives a descriptive benchmark of 5.35%. Forecasts become a buffered monthly throughput plan and expected turnover against the hotel’s 2.5-cycle Par Stock reference per category. The system is packaged as a local web application. The study demonstrates that domain semantics, temporal validation and operational interpretation matter more than model complexity in this short-series setting.
Download

Paper Nr: 103
Title:

From Quire to Query: An Ontology for Medieval Manuscript Knowledge

Authors:

Giulia Biagioni, Shari Boodts, Gleb Schmidt and Alex Wissink

Abstract: Manuscript libraries and archival repositories preserve collections of major cultural, historical, and scholarly value. However, information concerning medieval manuscripts is frequently dispersed across catalogues, databases, archival inventories, and scholarly publications, and is often represented through heterogeneous standards, terminologies, and levels of formalization. This fragmentation limits the accessibility, interoperability, and reuse of medieval manuscript knowledge across institutions. The Manuscript-AI project addresses this challenge through an agentic neuro-symbolic framework designed to transform heterogeneous manuscript information into interoperable semantic representations and knowledge graphs. At the core of this framework lies the Manuscript-AI ontology, a domain model developed to represent the textual, paratextual, codicological, palaeographical, material, historical, and provenance-related dimensions of medieval manuscripts. This paper presents both the ontology and the methodology adopted for its development. The process combined analysis of cultural heritage standards, review of manuscript studies literature, iterative collaboration with domain experts, competency-question driven modelling, and pilot-based validation. While aligned with the event-centric perspective of CIDOC CRM, the ontology adopts a structure adapted to the requirements of agentic neuro-symbolic workflows. The work contributes a semantic resource for manuscript libraries and a methodological approach for ontology engineering initiatives that must reconcile domain expressivity, interoperability, and agentic neuro-symbolic requirements.
Download

Paper Nr: 142
Title:

An Ontology for Transparent Risk-Driven Dependability Analysis of Cloud Services

Authors:

Giulia Biagioni and Claus Pahl

Abstract: Cloud services are commonly governed through service-level agreements (SLAs) that specify commitments such as availability, reliability, and compensation for violations. However, most SLA specifications focus on externally observable service levels and do not explicitly capture the technical hazards, risk factors, and mitigation measures that influence whether these commitments can be fulfilled. This limits systematic dependability-risk analysis across cloud-service lifecycles and federated service ecosystems. This paper presents OTRA-D (Ontology for Transparent Risk Analysis and Agreement for Dependability), an extension of the OTRA ontology for risk-driven dependability analysis of cloud services. OTRA-D preserves OTRA’s bow-tie representation of hazardous events, preconditions, postconditions, prevention measures, and remediation measures, while introducing explicit concepts for risk, probability, severity, and stakeholder impact. The ontology connects SLA-level commitments with technical conditions that may lead to service failures and their consequences for providers and consumers. OTRA-D is represented in OWL 2 DL and RDF(S), enabling machine-processable modelling and automated consistency checking. Its applicability is demonstrated through reliability, security, and availability scenarios, and through an analysis of the Microsoft Online Services SLA catalogue as an industrial reference framework. The results show that OTRA-D provides a semantic basis for linking SLA management with dependability-oriented risk assessment and mitigation.
Download

Short Papers
Paper Nr: 26
Title:

Generic Predicate Modeling for Knowledge Graph Quality: A Hybrid Semantic Filtering Approach

Authors:

Mussab Zneika

Abstract: Generic predicates such as labels, identifiers, and descriptive attributes are pervasive in knowledge graphs and often introduce structural noise that reduces interpretability and weakens structural discrimination. Existing filtering approaches mainly rely on heuristic or frequency-based strategies that fail to distinguish structurally informative relations from widely shared descriptive predicates. We propose Hybrid Semantic Filtering (HSF), a lightweight, data-driven, and algorithm-agnostic preprocessing framework that models predicate genericity using object uniqueness, predicate–class entropy, and relationality. We evaluate HSF on five real-world RDF datasets containing up to 6.1M triples using two representative RDF graph summarization methods and an RDF2Vec-based link prediction task. HSF consistently improves schema-level structural quality, achieving class precision gains of up to +0.49 (0.45 to 0.94 on LinkedMDB), while preserving instance-level semantics. It also improves relational embedding quality, increasing Macro-F1 from 0.76 to 0.84 on Jamendo while reducing graph size by approximately 20%. These results demonstrate that modeling predicate genericity improves both structural abstraction and downstream semantic representation learning across multiple graph analysis tasks.
Download

Paper Nr: 40
Title:

A Knowledge-Based Decision Support System for Explainable Retail Segmentation Using Clustering, RFM, and Semantic Modeling

Authors:

Vitor Vieira and David Aveiro

Abstract: In retail environments, the transformation of transactional data into actionable knowledge is essential for effective decision-making. This paper presents a knowledge-based Decision Support System (DSS) for retail analytics that combines unsupervised learning, explainable AI, and semantic modeling. Using Actuate Corporation’s ClassicModels dataset and SteelWheels as a case study, the proposed approach segments customers, products, and price sensitivity profiles through multi-dimensional feature engineering and clustering techniques, including K-Means, Hierarchical Clustering, DBSCAN, and Deep Clustering. Cluster quality is assessed through Silhouette, Calinski-Harabasz, and Davies-Bouldin metrics. To improve transparency, surrogate models based on Random Forest are used to explain cluster assignments with SHAP values. To better support knowledge engineering, the analytical outputs are formalized in a lightweight semantic layer that models core retail concepts such as Customer, Product, Segment, Discount Policy, and Recommendation, together with relations and rule-oriented interpretations derived from the explainability stage. This semantic representation enables the conversion of discovered patterns into explicit managerial knowledge, which is operationalized in Power BI to provide interpretable recommendations for retention, pricing, and catalog management. The study contributes an applied framework that connects data mining, explainability, and semantic knowledge representation for retail decision support.
Download

Paper Nr: 59
Title:

WaSO: An Ontology-Driven Knowledge Graph Framework for Smart Waste Sorting Digital Twins

Authors:

Aya Sahbi, Cheikh Brahim El Vaigh and Christophe Nicolle

Abstract: Modern waste sorting centers generate continuous streams of sensor events, equipment state transitions, and material quality measurements. Integrating and monitoring these heterogeneous data streams requires a formal, declarative framework-a requirement that is not fully addressed by existing industrial standards. To address this limitation, we introduce WaSO (Waste Sorting Ontology), an industrial-scale ontology designed to model the operational knowledge of smart sorting facilities. Developed in collaboration with industrial sorting centers and grounded in SCADA data, WaSO structures domain knowledge through an OWL 2 DL TBox coupled with a SHACL constraint layer to enforce data integrity. An automated instantiation pipeline then populates an ABox from industrial logs, thereby constructing a dynamic, large-scale knowledge graph. We validate the resulting knowledge base using four approaches: gold-standard instance evaluation, OWL DL consistency checking via Pellet, SPARQL competency questions covering the operational domains, and query evaluation through the OntoSQL pipeline. Ultimately, this multi-faceted evaluation demonstrates how hybrid validation strategies can certify the correctness and scalability of industrial knowledge graphs within Digital Twin architectures.
Download

Paper Nr: 64
Title:

Leveraging LLMs for Ontology-Grounded Symbolic Rule Generation in Context-Aware Automotive Applications

Authors:

Tianhao Gu, Adnan Bekan and Haonan Qiu

Abstract: Context-aware automotive applications rely on symbolic rules to derive situation-dependent behaviour. Authoring such rules is labour-intensive and error-prone because it requires understanding complex ontologies and translating informal requirements into correct logic. We present NL2SYM, which retrieves relevant ontology context, uses guided prompting and repeated sampling to generate candidate rules, and validates them through syntactic pre-screening and test-suite-based semantic validation, with conditional escalation to a stronger model after failure. On ten industrial automotive tasks, NL2SYM achieved a 79% overall semantic pass rate, providing initial evidence of feasibility.
Download

Paper Nr: 78
Title:

JSOnto: A Modular JSON-Based Syntax for Human-Friendly Ontology Engineering

Authors:

David Jilg and Ralph Bergmann

Abstract: We present JSOnto, a modular JSON-based ontology syntax for human-friendly, text-based knowledge engineering. Existing syntaxes and GUI editors are often difficult to edit manually, lack modularity, or do not scale well for large ontologies. JSOnto provides concise encodings of common RDF and OWL/OWL2 constructs, multi-file organization, inline annotations, and validation through JSON Schema. The Knowledge Transformation Framework supports validation and conversion of JSOnto to established syntaxes. JSOnto was developed in an iterative, user-centered process and evaluated in a within-subjects study with 26 participants comparing JSOnto, Turtle, and Protégé. Less experienced users preferred Protégé, whereas experienced users favored text-based modeling and preferred JSOnto over Turtle. This indicates that JSOnto is a practical complement to existing ontology tools, particularly for large-scale ontology development and maintenance.
Download

Paper Nr: 86
Title:

PIIRRITE: An Ontology for Inclusive Itineraries Modeling

Authors:

Salomon van Ginneken, Nadia Yacoubi Ayadi, Stéphanie Jean-Daubias and Gilles Gesquière

Abstract: Enabling inclusive pedestrian routing requires capturing users’ specific needs and preferences and leveraging environmental characteristics derived from diverse and heterogeneous data sources. In this paper, we introduce PIIRRITE (Pedestrian InclusIve indooR-outdooR ITinerariEs), an ontology designed to serve as a semantic backbone for inclusive pedestrian routing across indoor and outdoor environments. We build on existing domain ontologies such as GeoSPARQL to represent spatial entities and SAREF to attach contextual properties to spatial entities such as accessibility-related characteristics and environmental conditions. Our ontology is developed using the SAMOD methodology through successive modelets driven by a set of competency questions reflecting user requirements. By bringing together heterogeneous spatial and contextual data, the ontology enables querying and reasoning over diverse data sources to support personalized and inclusive itinerary planning. We demonstrate the applicability of our approach through a real-world case study based on the LyonTech La Doua campus.
Download

Paper Nr: 96
Title:

Arabic Legal NLP for Lebanese Judicial Decisions: Classification, Association Mining, and AI-Assisted Ruling Generation

Authors:

Mageda A.A. Sharafeddin, Adam Abul Karim, Zeinab Mouzaihem and Abbass Rammal

Abstract: This paper presents a three-stage knowledge-engineering framework for Arabic legal AI in the Lebanese judiciary. The study consolidates three primary Lebanese legal sources-CIJ rulings, Al-Adl journal decisions, and IDREL judge data-into a unified corpus of 12,349 rulings, supplemented by a local-court collection for the full-text full-text track. AraBERTv2 achieved 0.87 weighted F1 and 0.88 macro F1 for seven-class legal-area prediction and 0.91 weighted F1 for thirteen-class court prediction. Association mining using Apriori and Frequent Pattern (FP)-Growth produced identical rule sets and revealed interpretable topic–area and court– area patterns. For ruling generation, AraT5v2-base achieved ROUGE-L 0.61, BLEU 16.52, and multilingual cosine similarity 0.87 on 761 criminal-court examples. Judicial evaluation of the 77 test outputs found that 79% had either no difference or only a slight difference from the corresponding reference ruling, while legally important errors remained in a minority of cases. The three stages are evaluated independently and provide a foundation for future integration into an ontology-aware legal AI pipeline with formally defined and expert-validated knowledge representations.
Download

Paper Nr: 166
Title:

OQUO: Towards a Unified Semantic Model of Ontology Quality

Authors:

Francisco Abad-Navarro, Astrid Duque-Ramos, Manuel Quesada-Martínez, Belén Juanes-Cortés and Jesualdo Tomás Fernández-Breis

Abstract: Ontology quality assurance is essential to ensure that ontologies, as reusable and shareable knowledge artifacts, meet the standards required by the Semantic Web community and the FAIR principles. Existing quality frameworks propose diverse dimensions and quantitative metrics to evaluate ontologies, but they typically address only particular aspects of quality and lack a unified mechanism to harmonize their results. Several ontologies have been proposed to semantically represent quality evaluation elements, yet none covers all relevant aspects in an integrated manner. In this paper, we present the Ontology QUality Ontology (OQUO), a unified, framework-agnostic model that reuses and aligns these ontologies to provide a comprehensive semantic representation of ontology quality evaluation. OQUO supports the description of quality characteristics and metrics, measurement scales and their conversions, and the generation and classification of quality issues, enabling a holistic and comparable assessment without requiring manual harmonization of independent tools. We illustrate OQUO through use cases over the Gene Ontology and, most importantly, validate it in practice as the data model of QASAR, a deployed quality assurance platform in which OQUO makes the outputs of four independently developed frameworks (OQuaRE, HURON, OntoEnrich, and Evaluome) interoperable within a single model.
Download

Paper Nr: 167
Title:

Ontology-Grounded Entity Linking for Explainable Robot Perception: A Conceptual Framework

Authors:

Sorin Arion and Michaela Kümpel

Abstract: Autonomous robots operating in unfamiliar environments must do more than detect objects: they must connect perceptual observations to semantic object knowledge, reason about task-relevant roles, and communicate the basis of their decisions. This paper presents a conceptual framework for ontology-grounded explanations of robot perception in unfamiliar environments. An exchangeable perception front end produces segmented object candidates and open-vocabulary label hypotheses, which are linked to ontology concepts using lexical evidence, perceptual confidence, scene context, and task-specific constraints. When the available evidence does not justify a specific concept, the system performs hierarchical backoff and explicitly represents the remaining uncertainty. The resulting semantic Digital Twin retains the provenance of perceptual observations, entity-linking decisions, and ontology-based inferences. Task-scoped explanation graphs are then verbalised into natural-language descriptions of what the robot perceives, what it intends to do, and why a particular object or action was selected. A worked tabletop demonstration illustrates the framework’s entity-linking, uncertainty-handling and explanation capabilities.
Download

Paper Nr: 177
Title:

Binding Ontologies to Notation: A Declarative RDF Vocabulary for Notation-Independent Visualisation of RDFS/OWL Models

Authors:

Marek Suchánek and Jan Pecka

Abstract: Established tools for visualising RDFS/OWL ontologies (VOWL, Graffoo, OWLGrEd) each hard-code a single visual language: the mapping from ontological constructs to graphical elements is embedded in program code rather than exposed as a reusable artifact. A notation therefore cannot be shared, adapted to a domain, or moved between rendering engines without reimplementing the tool. Following a Design Science Research method, we developed VizBind, a declarative RDF vocabulary that describes a visual notation as data. It separates three concerns that prior approaches conflate: selection of model constructs via SPARQL, mapping to abstract visual constructs, and engine-agnostic styling, with priority-based conflict resolution. We evaluate the artifact along four questions. Of 16 VOWL constructs assessed, 9 are expressed in our subset and 5 more are expressible; the 2 that are not are procedural rather than mappings. Switching between a VOWL-subset and a UML-style notation changes only the notation document (no model bytes and no engine code), yet restructures the diagram from 17 nodes to 7. All 3 back-ends consume a byte-identical intermediate model, and both notations run warning-free over 6 ontologies including FOAF, DCAT, ORG, SKOS and PROV-O. Visual notations thus become configurations rather than tools.
Download

Paper Nr: 184
Title:

Investigating Automatic Ontology Generation of Biochemistry Experimental Procedures Using Large Language Models

Authors:

Tyler Johnson, Robert E. Mercer, Yetian Wang and Mohammed Alliheedi

Abstract: This position paper focuses on structuring the data encoded in biochemistry experimental procedures. The motivation arises from the need for reliable methods to ontologically represent procedural information. The central research question explores whether large language models (LLMs) can autonomously generate ontologies from procedural texts which are known to contain strong temporal and hierarchical components. The anticipated results from this research is a novel, automated ontology generation framework which shall include documented details of the challenges that procedural texts introduce. The first result shall be the novel construction of a ontological generation framework, and the second shall be a deeper investigation into the challenges of procedural texts. The anticipated outcomes will expand the scope of ontology generation to include procedural domains and will provide theory into the extraction and formatting of hierarchical and procedural textual structures. In particular, we have constructed a novel preprocessing algorithm and have documented an exploration into LLM prompting choices corresponding to our specific domain and have formulated an intermediate representation of latent entities.
Download

Paper Nr: 28
Title:

An Ontology-Based Framework for Explainable Situational Awareness in Connected Vehicles

Authors:

Itziar Urbieta, Ainhoa Lizaso, Marcos Nieto and Estíbaliz Loyo

Abstract: The transition toward Cooperative, Connected and Automated Mobility (CCAM) generates massive volumes of complex and heterogeneous data. Transforming this information into actionable knowledge is a fundamental Knowledge Engineering challenge, making effective data modeling crucial for the development of Intelligent Transportation Systems (ITS) and Connected and Automated Vehicles (CAVs). A key aspect of this evolution is fostering transparency and trust through Explainable AI (XAI) and cooperative perception enabled by Vehicle-to-Everything (V2X) communication. This paper addresses the need for a rigorous and scalable knowledge representation by introducing the Mobility Ontology for Situational Awareness (MOSA). MOSA formalizes the essential elements of traffic scenarios, focusing on modeling actions and events to enable semantic inference and enhance situational understanding. Supported by the construction of a Knowledge Graph (KG), the ontology facilitates critical event detection and the explainability of traffic situations in real-time. Finally, we present a service architecture that integrates MOSA for traffic event analysis and visualization, validated through a comprehensive simulation involving over 4,600 vehicles, serving as a practical, proven solution for both Traffic Management Centers (TMCs) and CAV users.
Download

Paper Nr: 53
Title:

OntoBinocular: Ontology-Based Access and Analysis of Student Software Projects

Authors:

Manuel Stöger, Michael Mayr, Johann Grabner, Mario Bernhart and Thomas Grechenig

Abstract: Software engineering (SE) project courses generate rich data from version control, issue tracking, and continuous integration tools, yet instructors struggle to monitor student teams effectively. Visualization-based Mining Software Repositories (MSR) tools, such as Binocular, support exploratory analysis through handcrafted views but keep it difficult to answer ad-hoc, semantically rich questions. We take the position that ontology-based access to MSR data complements visualization-driven tools by enabling declarative, cross-domain queries over a unified knowledge graph - a capability particularly valuable in SE education where instructors need flexible monitoring. To illustrate this potential, we present OntoBinocular, a semantic layer built on top of Binocular’s data model. OntoBinocular reuses and extends the established SemanGit and SEON ontologies and maps Binocular’s data exports to a unified RDF knowledge graph that supports inference and declarative querying. We demonstrate the approach through ten competency questions covering version control, issue tracking, and continuous integration, evaluated on a realistic student project from an SE course. Our findings suggest that a unified, queryable knowledge graph offers a promising direction for the flexible analysis of student project data, and we outline a roadmap for validating this position through broader evaluation.
Download

Paper Nr: 56
Title:

An Ontology for Megalithic Heritage

Authors:

Renata Vieira, Rafael Prezado, António Diniz, Alicia Núñez Garcia, Camila Campos, Ivo Santos, Leonor Rocha, Áurea Rodrigues, Saturnino Luz and Masood Masoodian

Abstract: This paper presents the Megalithic Heritage Ontology (MH Ontology), developed within the INT-ACT project, which explores the integration of tangible and intangible cultural heritage through immersive eXtended Reality (XR) experiences. The ontology is a specialization of the CIDOC Conceptual Reference Model (CIDOC-CRM) to represent the physical, symbolic and experiential dimensions of megalithic heritage sites. Two case studies, namely, Cromeleque das Fontainhas (Portugal) and Calanais I (Scotland), are used to demonstrate the ontology’s structure and application.
Download

Paper Nr: 73
Title:

AutoMeta: Automatic Metadata Generation for Tabular Databases with Profiling-Grounded Large Language Models

Authors:

Leandro Mendes Ferreira, Wesley Lourenco Barbosa, Maurício Adriano Fontes, Solange Nice Alves-Souza and Luciana Maria da Silva

Abstract: Metadata makes tabular databases reusable, searchable, and governable. Producing it by hand is slow, and an unconstrained large language model (LLM) may write fluent field descriptions that the data does not support. This paper presents AutoMeta, a reproducible pipeline that generates database metadata from statistical profiles and representative samples. It profiles each table, builds an evidence-rich prompt, and emits a data-dictionary record that must clear a strict schema behind a validate-and-retry gate. Quality is the cosine similarity between embeddings of the generated and reference descriptions. We evaluate the pipeline on a public benchmark of eleven relational databases, seventy-five tables, and seven hundred and ninety-eight reference fields, comparing six LLMs over 4,701 field-by-field comparisons under two embedding instruments. Grounded LLMs reach a mean similarity near 0.80 against the curated references. The most consistent model is deepseek-v4-flash: it wins the per-table head-to-head in 30 of 75 tables and is also the second cheapest. Difficulty depends as much on the data domain as on the model. The contribution is not a new technique but a controlled, auditable integration, measured end to end, that produces metadata artifacts for data catalogs without overclaiming full automation.
Download

Paper Nr: 85
Title:

Context: Towards Context-Aware Ontology Extension from Operational Metrics

Authors:

Hussain Hussain, Stefan Schöberl, Angelika Schneider and Verena Geist

Abstract: Organizations increasingly define operational metrics in structured, machine-readable formats to monitor systems, processes, and compliance. These metric definitions implicitly encode domain knowledge, such as referencing concepts, properties, and relationships, that often extends what is captured in formal ontologies. Yet the connection between operational metric catalogues and ontological knowledge remains manual, ad-hoc, and labor-intensive. We present COntExt, a framework for context-aware ontology extension that takes structured metric definitions as input and suggests how referenced concepts and properties should be integrated into an existing ontology, utilizing the context of these metrics. The framework defines the extension problem as three sub-tasks: parent class prediction, relation type prediction, and data property assignment. Across four cybersecurity ontologies, we evaluate different algorithms for each task. Our results show that metric-derived context improves the suggestions over ontology-context baselines for relation type prediction and data property assignment. Our work demonstrates that operational metric catalogues are a practical and underexploited source for ontology extension. This work enables organizations to maintain their ontologies at a significantly lower cost than manual engineering.
Download

Paper Nr: 111
Title:

JAR-KG: A Knowledge Graph for Assembly Sequence Optimization

Authors:

Sven Münker, Nenad Krdzavac, Julian Haller, Sylwia Olbrych, Felix Engel and Robert Schmitt

Abstract: Assembly sequence planning requires joint consideration of product structure, joining-process requirements, and precedence relations. Representing these aspects separately makes it difficult to query the relevant constraints together when assessing candidate assembly operations. This paper presents the Joining and Assembly Representation Knowledge Graph (JAR-KG), which integrates assembly structure and joining-process knowledge to make these constraints jointly queryable, providing a foundation for future simulation and optimization. The JAR-KG combines limited literature-based reconstructions of the Joining Process Selection (JPS) ontology and the National Institute of Standards and Technology (NIST) Assembly Representation (AR) ontology. A mapping pipeline computes five class equivalences to align shared vocabulary and checks class satisfiability at the schema level. An N-ary relation design pattern groups a connection feature, a candidate joining process, participating parts, geometry, and requirements within a shared context. This integration allows candidate joining processes to be represented and queried in relation to the parts and requirements of a specific connection. Individual assertions generated from CSV data for the Toy Dog assembly illustrate ontology population. Future work includes simulation and optimization, operation durations, resource requirements, a fact sheet, and comparative experimental validation.
Download

Paper Nr: 117
Title:

Plato: Generic Ontology-Based Representation of Semantic Planning Knowledge for Automatic PDDL Generation

Authors:

David Jilg, Alexander Schultheis and Ralph Bergmann

Abstract: Artificial Intelligence (AI) planning supports automated decision-making but requires explicit models, commonly maintained directly in the Planning Domain Definition Language (PDDL). Such models are difficult to create, reuse, evolve, and adapt to different planners. This paper presents Plato, a generic ontology schema that represents planning constructs as semantic objects. Plato serves as a maintainable source model from which the Knowledge Transformation Framework can generate planner-specific PDDL domain and problem files. A proof-of-concept evaluation covers classical, temporal, and numeric planning while also including automatic discretization. VAL validated the generated artifacts and plans, and off-the-shelf planners solved all tasks. The results provide initial evidence that Plato can serve as a semantic source model for automatic planning-domain generation.
Download

Paper Nr: 118
Title:

Pallas: Generic Asset Administration Shell-Based Representation of Planning Knowledge for Automatic PDDL Generation

Authors:

David Jilg, Felix Sehmer, Alexander Schultheis and Ralph Bergmann

Abstract: Artificial Intelligence (AI) planning supports automated decision-making, but difficulties in creating and maintaining formal planning models limit practical adoption. Planning knowledge is commonly encoded directly in the Planning Domain Definition Language (PDDL), coupling domain knowledge to planner-oriented syntax. Plato addressed this issue with an ontology-based semantic model. This paper presents Pallas, an Asset Administration Shell (AAS)-based counterpart to Plato for AAS-centered Industry 4.0 environments. Pallas represents planning concepts in standard AAS submodels and supports distribution across multiple AASs. The Knowledge Transformation Framework merges these models, applies planner-specific transformations, and generates PDDL. Four proof-of-concept domains and an exploratory user study provide initial feasibility evidence while revealing limitations of current AAS authoring tools.
Download

Paper Nr: 172
Title:

Multiple Autonomous Robot Ontology for Multi-Agent System Reasoning and Explainable Collaboration

Authors:

Alberto Olivares-Alarcos, Edison Pignaton de Freitas, Joanna I. Olszewska, Veera Ragavan Sampath Kumar, Amelie Gyrard, Syed Ashque MD Khaled, Maki Habib, Maksym Figat, Abdelghani Chibani, João Quintas, Howard Li and Paulo J. S. Gonçalves

Abstract: Autonomous, multi-robot systems (MRS) and multi-agent systems (MAS), which are based on symbolic reasoning planners and/or integrating neural Large Language Models (LLMs), demand interoperable, semantically rich frameworks for reliable, hallucination-free reasoning, explainable collaboration, and trustworthy interactions. Current standards, such as the IEEE 1872.2-2021 AuR ontology, offer foundational support but fall short in addressing the requirements of the adaptive, cloud-integrated, heterogeneous, multi-robot collaboration. This paper aims to outline the key limitations in existing standards and propose directions for the development of new ontologies such as the IEEE P1872.3 Multiple Autonomous Robot Ontology that aims to address the gaps of the 1872 standard series.
Download