KMIS 2026 Abstracts

Full Papers Short Papers
Full Papers
Paper Nr: 65
Title:

How Knowledge Management Drives Digital Transformation: The Differentiated Role of Organizational Learning
Authors:

Juan Ibujés-Villacís
Abstract: Digital transformation does not depend solely on the incorporation of technologies, but also on the organizational capacity to create, store, transfer, apply, and learn from knowledge. To examine this relationship, a quantitative approach was applied using confirmatory factor analysis and structural equation modeling. The knowledge management (KM) construct was structured based on five processes: knowledge creation, storage, transfer, application, and organizational learning. In turn, digital transformation (DT) was evaluated through indicators associated with digital innovation, digital knowledge infrastructure, collaborative platforms, data-driven decision-making, and digital culture. Information from 142 medium-sized manufacturing companies in the province of Pichincha, Ecuador, was used for the study. The findings indicate a positive and statistically significant relationship between KM and DT. However, the specific model suggested that the associations between each KM process and DT indicators were not homogeneous. Organizational learning was the only process significantly associated with several DT indicators, particularly digital innovation, data-driven decision-making, and digital culture. In contrast, knowledge creation, storage, transfer, and application did not show significant independent associations in the specific model. These findings suggest that, in this sample, the relationship between KM and DT is especially evident when KM is linked to organizational learning, adaptation, and the institutionalization of new organizational practices.
Download
Paper Nr: 68
Title:

Architectural Components for Dengue Data Integration: An Approach Aligned with the ISO/IEC 20547-3 Big Data Reference Architecture
Authors:

Wesley Lourenco Barbosa, Leandro Mendes Ferreira, Solange Nice Alves-Souza, Francisco Chiaravalloti-Neto and Jorge Luis Risco Becerra
Abstract: The global expansion of dengue and other arboviral diseases poses a critical challenge to public health systems, requiring timely and evidence-based responses. In this context, the adoption of Big Data platforms has emerged as an essential strategy for integrating large volumes of heterogeneous data and supporting decision-making processes. However, existing technological platforms still face limitations, including fragmented data infrastructures, a lack of standardization, and the absence of granular information capable of addressing the operational needs of local epidemiological surveillance. Hence, this study proposes a set of architectural components for dengue data integration, guided by the principles and recommendations of the ISO/IEC 20547-3 Big Data Reference Architecture (BDRA). The methodology comprised requirements elicitation with municipal health managers and adaptation of the BDRA functional layers into seven operational logical components. The functional decomposition of the architecture, combined with the granular mapping of roles across multidisciplinary teams, enabled a clear separation of responsibilities and supported the establishment of effective governance processes. The resulting architecture provides a flexible, modular, and reproducible framework capable of supporting local public health decision-making and facilitating rapid responses to dengue outbreaks. Furthermore, its extensible design allows its application to the monitoring and surveillance of other urban arboviral diseases.
Download
Paper Nr: 75
Title:

SEPath: A Self-Evolving Framework for Auditable Financial Reasoning via KG-Guided Adaptive CoT and Constrained ReAct
Authors:

Tsung-Nan Chou
Abstract: The Large Language Model (LLM) integration has revolutionized data access across critical sectors like finance. However, applying generalist models to high-stakes domains presents severe challenges regarding compliance, verifiable provenance, and procedural fidelity. Current architectures, including standard Retrieval-Augmented Generation (RAG), unconstrained Chain-of-Thought (CoT), and basic ReAct loops, fail to guarantee the structural rigor and deterministic execution required by regulated industries, leaving gaps in action validation and systemic governance. To address this, we propose SEPath (Structured Expert Path), a cognitive control architecture for auditable financial reasoning. Instead of treating reasoning as an open-ended generative process, SEPath models it as a constrained, rule-bound state machine traversal. Governance is achieved via three core mechanisms: (1) a Financial Logic Graph (FLG) transforming static Knowledge Graphs into executable blueprints encoding prerequisites and causality; (2) a specialized CoT Selector to dynamically mandate professional thinking strategies; and (3) strict execution enforcement via a Constrained ReAct loop. Critically, SEPath incorporates an Active Self-Evaluation mechanism calculating a composite Feedback Reward Score. This allows the system to interpret failure modes (e.g., procedural gaps or factual inconsistencies) and autonomously update both the FLG structural knowledge and reasoning templates. Continuous learning evaluations demonstrate SEPath's capacity for self-improvement, including recovering from performance dips and correcting internal constraints. By binding structural knowledge (FLG), controlled methodology (CoT Selector), and compliance enforcement (CR-ReAct) into a unified, auditable pipeline, this work establishes a new approach for trustworthy AI in mission-critical financial environments.
Download
Paper Nr: 133
Title:

Toward Trustworthy Local RAG for Institutional Document Consultation
Authors:

Davi Soares, Carlos Eduardo Barbosa, Matheus Argôlo, Lucas Nóbrega, Luiz Felipe Martinez and Jano Moreira de Souza
Abstract: Institutional organizations rely on large technical, regulatory, and administrative document repositories for decision-making and compliance analysis. These repositories are often dispersed across heterogeneous systems and remain difficult to query through conventional keyword-based search. Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) enable conversational access to such knowledge, but their adoption in privacy-sensitive environments requires stronger control over grounding, scope, and data exposure. This work proposes a local RAG workflow for institutional document consultation, designed to operate without transmitting documents or user queries to third-party services. The workflow integrates document ingestion, hybrid lexical-semantic retrieval, context-grounded generation, independent classifier-based guardrails, and feedback-driven prompt adaptation within a unified architecture. The workflow was instantiated and analyzed on a heterogeneous institutional repository containing technical, regulatory, and academic documents, to investigate the operational behavior of trustworthy locally executed RAG pipelines under different consultation regimes. The results illustrate three behaviors: grounded response generation when reliable context is available, explicit refusal when retrieval does not provide sufficient evidence, and early blocking of out-of-scope interactions. Explicit refusal regimes combined with independent guardrails and hybrid retrieval can reduce unsupported generation while preserving modularity and local execution in privacy-sensitive environments.
Download
Paper Nr: 164
Title:

AI-Driven Digital Transformation: Challenges, Implications and Research Themes from Insights of Swedish IT Consultants
Authors:

Victoria Klyukina, Maximilian Langselius Comstock and Vilhelm Hedin
Abstract: Realising digital transformation (DT) using artificial intelligence (AI) requires coordination between business, legal, technical, and governance functions. This raises questions about how to organise and manage strategic AI-driven DT. Although, the topic is continuously investigated in the recent years from multiple theoretical perspectives, the empirical investigation is rather scarce, and to large extent takes quantitative approach. This paper presents practical challenges associated with AI-driven DT taking a qualitative approach with data collection in a Swedish IT consulting organisation. The analysis of empirical data resulted in explication of four practical challenges, three implications for business development and six emerging research themes. The study contributes to deeper understanding of the practical challenges associated with AI-driven DT, and the results can be used as a research agenda to assist researchers, practitioners and policymakers.
Download
Paper Nr: 175
Title:

Understanding the Challenges for AI-Supported Guidance for Free Trade
Authors:

Philip Klinger, Tim Klinger, Tolga Yilmaz, Lana Sattelmaier and Jan Pawlowski
Abstract: Small and medium-sized enterprises (SMEs) seeking international market access face complex regulatory requirements that are difficult to interpret and translate into action. This study examines how digital support systems can facilitate regulatory learning and competence development for Indonesian SMEs in the context of the EU–Indonesia Comprehensive Economic Partnership Agreement (IEU-CEPA). Following a Design Science Research approach, the study focuses on the problem analysis and combines a literature review with expert interviews. Nine expert interviews and one focus group were conducted with representatives of export-oriented SMEs and SME-related experts from Indonesia. The findings show that the main challenge is not access to information, but transforming regulatory knowledge into actionable competence. Based on these findings, a validated problem statement is formulated.
Download
Short Papers
Paper Nr: 38
Title:

“What You Should Know about AI”: Evaluating a Virtual Reality Demonstrator for AI Knowledge Transfer in the Workplace
Authors:

Verena Pohl, Selina Layer, Nadine Lahn, Felix Scheerer, Janina Weber and Emma Weygoldt
Abstract: AI literacy is increasingly essential as AI adoption transforms workplace practices and human-AI collaboration. However, interactive training formats with game-like elements for AI beginners remain scarce. This paper addresses the question: Is virtual reality (VR) an effective tool for transferring AI knowledge to novices? A VR demonstrator was developed, designed to convey fundamental AI concepts – including machine learning and neural networks. “What you should know about AI” allows users to explore AI-related content through interactive, gamified, and multimedia-rich virtual environments. The demonstrator was evaluated with 27 participants through pre- and post-quizzes, usability assessments, and interviews. Results showed significant improvement in AI knowledge after VR exposure. Participants with greater VR controller experience achieved higher post-quiz scores, while general VR experience and age showed no significant correlation. Participants reported high immersion but also occasional challenges with VR handling and information processing. The use of VR for educational purposes was generally viewed positively. Findings suggest that supervised VR demonstrators can effectively introduce AI concepts to non-experts in business contexts, especially when users receive clear instruction and technical support.
Download
Paper Nr: 42
Title:

A Cloud-Based Architecture for Knowledge-Driven Battery Management across First and Second Life Applications
Authors:

Rui Abreu, Nuno Marques and Marco Rodrigues
Abstract: The rapid expansion of battery energy storage systems across mobility, industrial, and grid applications has generated large volumes of heterogeneous operational data, challenging traditional Battery Management Systems (BMS) in terms of scalability, interoperability, and knowledge reuse. This paper proposes a cloud-based, knowledge-driven architecture that explicitly structures the lifecycle of battery-related knowledge, spanning acquisition, ingestion, persistence, processing, generation, and exploitation. The architecture integrates IoT technologies, scalable cloud services, and machine learning pipelines to enable continuous knowledge extraction from distributed battery assets across both first and second life applications. A key contribution lies in the explicit treatment of knowledge as a first-class entity, supporting traceability, explainability, and lifecycle-aware reuse, including second-life decision support. The architecture incorporates a knowledge-centric security and governance model ensuring controlled multi-stakeholder access. It also considers a use-case by instantiating Microsoft Azure services and validating in a demonstration environment, highlighting its ability to support scalable, secure, and interoperable battery analytics and knowledge management.
Download
Paper Nr: 45
Title:

Knowledge Management in Travel Agencies: A Bibliometric Analysis
Authors:

Nataša Rupčić, M. Bahadır Kalipçi and Erkan Kadir Şimşek
Abstract: This study presents a comprehensive bibliometric analysis of knowledge management research in travel agencies, examining 97 papers published from 1990 to 2026. The findings clarify the discipline’s developmental trajectory, highlighting its shift from early discussions of human resource management to the emergence of specialized, tourism-focused knowledge management frameworks in the mid-2000s. This trend reflects the field’s increasing conceptual sophistication and contextual adaptation. The analysis identifies key thematic clusters that have shaped the literature, including knowledge sharing and transfer, innovation, digital technology adoption, and customer satisfaction. It underscores the alignment between theoretical developments and the practical challenges faced by travel agencies in an increasingly digital environment. The geographic distribution of research output shows a concentration in developed economies, particularly the United States and China, along with notable patterns of international collaboration. This study contributes to the literature by offering a systematic review of the interplay between knowledge management and tourism agencies, highlighting dominant research streams and suggesting potential for future research.
Download
Paper Nr: 51
Title:

A Mixed-Method Study of Agentic AI Adoption Capabilities and Decision Transformation in European Organizations
Authors:

Domenico Pochiero, Hans Mulder and Steven De Haes
Abstract: Agentic AI systems are a step beyond decision-support AI, where AI acts as an agent in an autonomous way for multiple steps. But companies are dealing with a lot of uncertainty regarding their responsible adoption and integration. The paper examines the organizational, technological, and institutional aspects influencing the adoption of Agentic AI in European organizations by conducting a mixed-method Design Science Research study that includes qualitative interviews with 4 agentic AI practitioners and a quantitative survey of 302 professionals from various sectors. Interviews focused on senior practitioners within the aviation, insurance, technology, and consulting sectors, who have strategic oversight over AI governance and at least five years of experience with AI. The results indicate that adoption is not a function of one factor, but a combination of organizational capabilities like leadership readiness, culture for innovation, the maturity of the data, and the formality of the governance. The structural tension between ROI orientation and innovation culture suggests that the logic of financial governance could limit the use of autonomous exploration of AI. An Agentic AI Readiness and Integration Framework is proposed as a prescriptive design artifact to help organizations evaluate readiness and plan for adoption in terms of five dimensions: organizational capabilities, technological readiness, institutional context, human factors, and the agentic dimensions of autonomy, delegation, and orchestration.
Download
Paper Nr: 69
Title:

Enhancing Immigrant Integration through Human-AI Adaptive Learning and Digital Skills Development
Authors:

Azadeh Sarkheyli and Arezoo Sarkheyli-Hägele
Abstract: Highly qualified migrants, particularly women, face persistent barriers to labour-market integration, including limited recognition of prior learning, language-related challenges, and fragmented adult education pathways. At the same time, municipalities and employers are experiencing increasing demands for digital competence and skilled labour. This research proposes a personalised, digitally supported experiential learning model aimed at supporting more effective transitions into employment and further education. The study builds on research in experiential learning, digital inclusion, and human-centred educational technologies, and introduces a human-AI adaptive learning approach designed to support individual learning trajectories and labour-market readiness. A mixed-methods research design is proposed, combining qualitative and quantitative approaches, including interviews, observations, co-design workshops, usability studies, learning analytics, and administrative data analysis. Pilot implementations in selected schools are planned to support iterative development and evaluation of the proposed model. The research aims to contribute knowledge on the design and implementation of adaptive adult education services that promote digital inclusion, labour-market integration, and more sustainable municipal welfare practices.
Download
Paper Nr: 72
Title:

Enablers of and Barriers to Knowledge-Based Management: Empirical Insights from Public Sector Organisations
Authors:

Emma Kivimäki and Nina Helander
Abstract: Knowledge-based management (KBM) supports systematic and holistic decision-making in organisations. Although Finnish public sector organisations have increasingly invested in KBM development, understanding of the factors influencing its implementation remains limited. This article identifies enablers of and barriers to KBM through interviews with 16 individuals responsible for KBM development in public sector organisations. Qualitative content analysis revealed four categories for enabling or hindering factors: technology, data, information and knowledge (DIK), culture and structures. The findings indicate that successful KBM requires balanced development of all four categories with knowledge-based culture and technical capabilities emerging as particularly important enablers. Furthermore, the results highlight DIK as a distinct category influencing KBM and suggest a need for greater attention to DIK-related enablers and barriers in future research. The study contributes to knowledge management research on enablers and barriers by providing rich empirical insights from the perspective of KBM.
Download
Paper Nr: 76
Title:

Automation as Support for the Effective Implementation of Data Governance: A Literature Review
Authors:

Carlos Alberto Bassi, Solange Nice Alves-Souza and Luiz Sergio de Souza
Abstract: Implementing Data Governance (DG) requires managing complex interactions across data assets, business processes, organizational policies, and corporate culture. In this context, automating DG processes and activities can mitigate human error and improve operational continuity. This Literature Review analyzes 30 case studies to investigate which approaches and techniques were adopted, where they were applied, and the challenges and benefits involved. It was possible to identify 13 distinct automation approaches and techniques, mapped across 15 functional applications categorized into 4 domains: Data, Content, Process, and Organizational. Findings reveal that while automation substantially improves data quality, compliance, and processing speed, its implementation faces critical trade-offs regarding computational overhead, system integration complexities, and data privacy risks. This study clarifies the practical capabilities and boundaries, providing an evidence-based foundation to build and implement a solution to automate DG tasks and processes.
Download
Paper Nr: 90
Title:

CURACT: A Curation-Anchored Framework for Generative-AI Report Writing in Knowledge-Intensive Expert Work
Authors:

Janne Kauttonen, Umair Ali Khan and Dmitry Kudryavtsev
Abstract: Professional report writing in knowledge-intensive organizations often requires experts to combine private, heterogeneous data, including documents, notes, recordings, photographs, and structured records. Such reports must follow templates, remain traceable to evidence, and be accountable to human experts. Existing long-form generative AI systems mainly target open-web article writing, while generic and agentic retrieval-augmented generation systems do not specify how private multimodal evidence, tacit expert knowledge, human review, and evidence traceability should be organized for enterprise reporting. This position paper proposes CURACT, a Curation-Anchored, Governed, Traceable Reporting framework grounded in knowledgemanagement process models, the SECI model, and its GenAI-era extension through GRAI. CURACT divides report writing into three stages: multimodal data ingestion and normalization, section-bound knowledge curation, and governed synthesis. Its main contribution is a persisted curation layer in which factual units are organized by report section before prose generation. Source hierarchy, role separation, human checkpoints, and uncertainty marking support this layer. The framework is illustrated through a real industrial use case of building-condition-assessment reporting and implemented as a generic, open-source toolkit module.
Download
Paper Nr: 93
Title:

From Task-Level Gains to Firm-Level Performance under Generative AI: A Synthesis for Knowledge Management and Information Systems
Authors:

Anton Ludwig Bonin, Daniel Staegemann, Christian Haertel, Matthias Pohl and Klaus Turowski
Abstract: At the individual level, productivity effects are in the double digits. In recent studies, agents in customer support with access to a conversational assistant resolved more issues per hour, college-educated professionals took far less time on writing tasks and software developers who adopted an AI coding assistant completed more tasks per week. Consultants did better on a task inside the reach of the AI and worse on one built to fall outside it. At the firm level, the evidence is less plentiful and is far less conclusive. Why these results might differ at larger scale has rarely been discussed in one place. The paper is a narrative synthesis and it reviews the available evidence and discusses four lines of work that each consider the gap from a different side: coordination, capability configuration, deployment modes and workforce psychology. While each area of literature has its own vocabulary to describe similar concepts, the areas have not converged or connected with one another. We argue that the discrepancy between the individual task and firm levels is unlikely to be due to measurement error alone and may reflect how an uncertain external language-based tool interacts with a knowledge-rich organization.
Download
Paper Nr: 100
Title:

Knowledge Graphs for Engineering Competency Upskilling in Large Enterprises
Authors:

Saurabh Jain, Sandesh Pai and Juliette Mattioli
Abstract: Large industrial enterprises in aerospace, defense, energy, automotive, and advanced manufacturing depend on their engineering workforce as their most knowledge-intensive asset, yet the tools used to model competencies remain largely static. This paper argues that knowledge graphs (KGs) provide an appropriate representational and reasoning substrate for competency management in such settings. We formalize a role–competency– mission ontology grounded in OWL 2, define a competency gap function for dynamic gap analysis, and characterize the structural advantages of KG-based representations over flat skill taxonomies. We then present the Spire.AI Large Graph Model (LGM) for Skills, a domain-aware KG comprising over fourteen million skill nodes with a self-evolving ingestion pipeline, describe its implementation stack and an end-to-end gap-analysis workflow, and propose a four-dimensional knowledge quality assessment framework spanning completeness, accuracy, timeliness, and consistency. The discussion situates the approach within the skills-based organization paradigm and the neuro-symbolic AI tradition, and outlines open challenges in cold-start handling, ontology alignment, and explainability via Graph-RAG.
Download
Paper Nr: 105
Title:

Pricing Artificial Intelligence Tools: A Value-Informed Perspective
Authors:

Tommi Mahlamäki, Heini Kinanen and Jussi Myllärniemi
Abstract: Artificial intelligence (AI) tools are increasingly embedded in software products, knowledge-intensive workflows, and decision-making environments. They are expected to automate tasks and improve work efficiency. Yet pricing remains difficult because customer value is context-dependent and often visible only through use. This conceptual paper examines the aspects that should be considered when pricing AI tools from a value-informed perspective. Drawing on literature, it develops a three-category framework for pricing AI tools. The framework includes customer value potential, value realization conditions, and pricing and value capture. It links customer use context and stakeholder value with readiness conditions and value-capture mechanisms. The paper argues that pricing AI tools should be guided by more than technological novelty or supplier costs. Pricing should combine value-informed logic with cost- and competition-informed reference points. The contribution is a category-based conceptual framework for pricing AI tools in business-to-business (B2B) software and digital solution contexts.
Download
Paper Nr: 107
Title:

Designing a Customer-Centric Operating Model Framework for AI-Readiness
Authors:

Damian Copeland and Kathy Mogford
Abstract: As artificial intelligence becomes ubiquitous across businesses combining professional and managed services, traditional operating models fail to support the fluid knowledge sharing required for continuous system evolution. Because large language models (LLMs) are inherently non-deterministic, deployment is no longer a finite project phase; instead, it demands ongoing lifecycle management. This shift redefines technical roles, moving specialised resources from front-end development to operational functions focused on continuous model monitoring, triaging, and retraining. Consequently, traditional silos between professional and managed services are no longer viable, requiring dynamic resource allocation and integrated knowledge structures. Concurrently, clients require a single point of contact capable of identifying portfolio solutions to deliver business outcomes; a capability exceeding traditional account management boundaries. Adopting a Design Science Research (DSR) approach, this paper develops an integrated operating model framework designed to resolve these cross-functional knowledge and organisational bottlenecks. Grounded in an illustrative scenario, we show how to structure a unified service delivery function, then extend the framework with further design principles ready for validation within an organisational context.
Download
Paper Nr: 130
Title:

Knowledge Sharing in Strategic Knowledge Management in a Finnish Social and Healthcare Organisation: Barriers and Benefits
Authors:

Nina Helander and Krishna Venkitachalam
Abstract: The world’s modern societies are significantly shaped by public sector organizations. Consequently, the creation and sustenance of valuable knowledge are paramount in such organizations. To effectively manage organizational knowledge, consistent and coherent knowledge sharing is essential. Organizations may encounter various problems without knowledge sharing, including reduced employee well-being and performance. This paper seeks to explore the barriers and benefits of knowledge sharing within the Finnish public sector. The empirical data comes from one case study concerning a substantial public social and healthcare organization located in Finland. Analysis of the empirical data suggests that this research adds to the literature by offering real-world observations on knowledge sharing as a key aspect of strategic knowledge management within the public sector. The study addresses call for context-specific and practice-oriented research in public sector by examining both the barriers and benefits of knowledge sharing within one organisation. Our findings contribute to a better understanding of how knowledge sharing impacts strategic knowledge management in the context of public sector and presents some practical propositions for developing knowledge sharing in public organisations.
Download
Paper Nr: 136
Title:

Longitudinal Analysis of Federal Sanitation Transfers and Municipal Structural Evolution in Brazil
Authors:

Davi Soares, Carlos Eduardo Barbosa, Matheus Argôlo, Lucas Nóbrega, Luiz Felipe Martinez and Jano Moreira de Souza
Abstract: Federal sanitation transfers generate large volumes of open governmental data regarding agreements, financial execution, and municipal infrastructure conditions. However, these datasets are typically analyzed separately, limiting the ability to describe federal investments jointly with observed sanitation evolution over time. This work proposes an exploratory screening framework for longitudinal territorial integration, designed to flag municipalities with high observed improvement under substantial federal disbursement, as well as high-disbursement municipalities with limited observed improvement. The framework integrates federal agreements data and Demographic Census indicators under a unified Brazilian Institute of Geography and Statistics (IBGE) territorial identifier. A composite sanitation deficit is constructed from three structural dimensions: adequate sanitation coverage, water supply coverage, and prevalence of rudimentary septic systems. Based on this deficit and current cumulative federal disbursement associated with agreements signed during the intercensal interval, the framework generates two complementary rankings under explicit attribution filters. It was instantiated over sanitation agreements historically associated with the National Health Foundation (FU-NASA), covering 5,350 qualified municipalities and more than 8,000 agreements signed between the 2010 and 2022 Demographic Censuses. The results reveal substantial heterogeneity between municipalities regarding the joint distribution of federal disbursement and observed structural evolution. The design is observational and provides no counterfactual: the rankings support prioritisation for further inspection, and neither establish nor rule out an effect of the transfers. The work contributes a formally specified method for longitudinal territorial integration of open governmental datasets.
Download
Paper Nr: 145
Title:

Towards Explainable LLM-Based Decision Support for Requirements Traceability Verification
Authors:

Christophe Ponsard
Abstract: Requirements traceability captures engineering knowledge linking stakeholder intentions to requirements, design, and test artefacts throughout the system lifecycle. In large industrial projects, maintaining this traceability is costly and largely manual, as engineers must preserve relationships in an evolving context. Incomplete or inconsistent traceability hampers verification and certification and leads to the loss of valuable knowledge for design rationale, impact analysis, and reuse. Recent advances in Large Language Models (LLMs) and agentic AI systems offer new opportunities to automate the acquisition, validation, and maintenance of traceability knowledge through natural language understanding and reasoning. This short paper presents an AI-assisted approach to traceability knowledge management focused on verifying semantic consistency between requirements at successive refinement stages. Building upon theoretical and practical models of requirements refinement, including goal-oriented methods, we derive a set of knowledge validation rules that are implemented through an LLM-powered workflow easing decision support by the analyst in charge. Our work seeks to balance accurate identification of OK vs NOT OK cases with the ability to flag issues that require MANUAL review, while also providing visualisation and explainability to support deep understanding and interpretation of the results. Our work is evaluated on public traceability benchmarks and an anonymised industrial case study. To challenge the detection capabilities, we also rely on mutation operators on valid traceability.
Download
Paper Nr: 155
Title:

When Artificial Intelligence Becomes Knowledge Orchestration as Organizational Ambidexterity
Authors:

Nabil Georges Badr
Abstract: Artificial intelligence is reshaping how organizations create, integrate, and steward knowledge, dissolving structural separations between exploration and exploitation. Through a scoping review of 92 studies, this paper conceptualizes AI-enabled ambidexterity as a distributed sociotechnical capability model spanning computational, organizational, and governance intelligence. Computational systems generate and refine knowledge; organizational mechanisms translate and diffuse it; governance structures ensure transparency, accountability, and responsible use. These interact to form a unified knowledge-orchestration ecosystem that redistributes agency across humans and machines. The paper clarifies boundary conditions for AI-enabled knowledge systems and offers recommendations for ambidextrous knowledge orchestration.
Paper Nr: 160
Title:

Cyber Influence and Systemic Vulnerabilities in Digital Social Care Ecosystems
Authors:

Ilkka Tikanmäki
Abstract: Social services have been digitalised, resulting in a rise in the risk of cyber threats to individuals, organisations, and society. Cybercriminals are exploiting structural vulnerabilities in social care organisations, which include sensitive customer data and dependence on shared information systems. The purpose of this study is to investigate how social services are affected by cybersecurity risks due to technology, system dependencies, and human factors, particularly in the Nordic countries. Three key findings emerge. 1) excessive access rights and internal security breaches are the most significant recurring risks, 2) systemic vulnerabilities can lead to the collapse of entire service chains when relying on shared systems and single suppliers and 3) suppliers and back-end systems are often the source of cyberattacks on social services, rather than solely through social work systems. EU-level initiatives have strengthened cybersecurity protection in healthcare; the social care sector remains comparatively vulnerable. Improving resilience and secure social services in an increasingly hostile digital environment requires strengthening governance, access control, supply chain security, and local cyber capabilities. Social sector organisations and societies can suffer a significant impact from cyber-attacks by disrupting people's daily lives. Supply chains can be damaged and affected by attacks depending on their sophistication.
Download
Paper Nr: 161
Title:

Mapping IT Skills Trends: An Analysis of LinkedIn Job Listings
Authors:

Luan Vieira, Lucas Nóbrega, Luiz Felipe Martinez, Carlos Eduardo Barbosa, Matheus Argôlo and Jano Moreira de Souza
Abstract: The rapid evolution of the Information Technology (IT) sector demands continuous adaptation by professionals and organizations, yet labor-market information is frequently available as fragmented and unstructured text. This study examines how skills and salary information can be extracted and standardized from LinkedIn job postings to support the analysis of competency and compensation patterns. We operationalized an automated workflow combining web scraping, Large Language Model (LLM)-based information extraction, skill normalization, salary standardization, association measures, and clustering. The analysis was conducted on 6,739 postings representing thirteen predefined IT roles, of which 1,857 contained identifiable salary information. Within this cross-sectional sample, Python, SQL, and AWS were the most frequently identified skills, while skills related to machine learning, cloud computing, and specialized infrastructure tended to occur in postings with higher salary midpoints. These associations are descriptive and should not be interpreted as causal effects or representative estimates of the entire IT labor market. The contribution is primarily integrative and empirical, providing a reproducible workflow for structuring unstructured labor-market information and generating knowledge that may inform career development and organizational competency planning.
Download
Paper Nr: 37
Title:

NetworkBPM: Diagnosing Tasks and Operational Knowledge Risks in Business Processes
Authors:

Anna Ujwary-Gil and Anna Florek-Paszkowska
Abstract: This paper presents NetworkBPM, a two-mode network diagnostic for locating operational knowledge risk in business processes. The method links a binary Actor × Task matrix to a Task × Business Process matrix and derives weighted actor-process coverage while retaining task-level traceability. Unlike aggregate responsibility tables or actor centrality rankings, NetworkBPM measures exact task redundancy, shallow process coverage, and the effect of excluding individual actors. An illustrative application to an anonymized information-technology organization contains 45 actors, 48 tasks, seven processes, and 708 actor-task ties. 249 of 315 actor-process pairs are positive (density 0.790), four tasks depend on one actor, one task depends on two actors, and 54.6% of positive actor-process ties cover only one or two tasks. BP01 is connected to 39 actors yet contains three single-actor tasks, and removing A22 would uncover all three. The contribution is a reproducible, process-specific operationalization of knowledge concentration that complements business process management (BPM), organizational mining, and organizational network analysis when event logs or competence measurements are unavailable. The method supports targeted cross-training and documentation but is a structural screening tool, not a measure of workload or expertise quality.
Download
Paper Nr: 63
Title:

Towards Privacy-Compliant Healthcare Data: Developing an Efficient Multilingual Filter for PII and PHI under Safe Harbor Principles
Authors:

Andrin Benz, Jannis Schönleber, Bella Fernández and Bernhard Bermeitinger
Abstract: Medical text anonymization is essential for data-driven healthcare but constrained by strict privacy regulations, particularly in multilingual Swiss settings. A lightweight, privacy-compliant filter for removing personally identifiable and protected health information from unstructured medical text is presented. The system combines regular expressions, rule-based logic, and locally-run guided LLMs, where local open-source models are constrained to assign a label to each word. This approach is evaluated on 50 manually annotated SwissGerman medical documents and 520 English i2b2 notes. A regex-only baseline shows substantially lower F1 score and higher leakage on both datasets. Overall, the proposed hybrid approach reduces residual identifiers while maintaining deterministic runtime and modest computational cost, enabling practical medical text anonymization in multilingual contexts.
Download
Paper Nr: 70
Title:

An Education Domain Model for Semantic Interoperability and Continuity of Education
Authors:

Kristina Murtazin, Oleg Shvets, Karl-Erik Karu, Mart Roost, Viljam Puusep and Gunnar Piho
Abstract: Education lacks a standardised conceptual framework for semantic interoperability and continuity comparable to the continuity-oriented concept systems used in healthcare. This paper presents an Education Domain Model developed by systematically transforming ISO 13940 ContSys concepts, relations, and process logic for education. The model covers educational actors, matters, activities, processes, planning, responsibilities, records, governance, and learner lifecycle management. Its scope is the common conceptual layer shared by different forms of education; it does not assume that schools, universities, vocational education, and professional training have identical goals or methods. The main structural change is role-centric actor modelling, which separates a person or organisation from the role through which it participates in education. The study defines explicit transformation rules, compares the model with the European Learning Model (ELM), European Digital Credentials for Learning (EDC), and EMREX (a digital solution for the exchange of verified student records across borders), and reports on internal semantic, structural, and expert reviews. The results establish conceptual feasibility, but operational interoperability and continuity remain to be validated through implementation and external evaluation.
Download
Paper Nr: 94
Title:

Spending Is Not Capability: Why Generative AI Investment Alone Does Not Yield Firm-Level Returns
Authors:

Anton Ludwig Bonin, Daniel Staegemann, Christian Haertel, Matthias Pohl and Klaus Turowski
Abstract: Spending on Generative AI tools has grown quickly, while the returns that firms report remain low relative to their expectations. This conceptual paper examines the gap through the resource-based view, which speaks to relative competitive position, and argues that the gap is more than a temporary adoption lag. A foundation model is typically rented from an external provider on terms broadly open to any competitor, so access to it places a firm on much the same footing as its rivals. A sustained advantage requires at least three assets that competitors cannot simply lease: proprietary data, work processes redesigned around the model itself and evaluation infrastructure for probabilistic output. Building these complementary assets is expensive and slow, which is consistent with the productivity J-curve and with an industry study in which most firms showed no measurable return. On this basis, the paper describes what knowledge management and information systems research should measure.
Download
Paper Nr: 95
Title:

Audio Bookmarking and Text Extraction System for Multi-Domain Knowledge Management
Authors:

Dara Pir
Abstract: This position paper presents the design of a lightweight, user-centered audio tool for efficient extraction of text from segments of recorded speech. The system provides a common platform that can be readily adapted and enhanced for various application domains, while serving as the basis for planned user studies to further refine its utility. It enables users to play an audio file, navigate through it, and designate specific points of interest with timestamped bookmarks for future reference. These bookmarks can then be used to transcribe the associated spoken words into text via automatic speech recognition technology. Additionally, users can export their collection of extracted text to an open format, allowing the information to be incorporated into existing workflows and knowledge management systems. The proposed design is illustrated through screenshots of a prototype, demonstrating how the different components integrate to provide an intuitive and seamless user experience. The features and core interface design of this system provide a foundation to support the development of specialized, user-driven AI tools across its diverse application domains.
Download
Paper Nr: 114
Title:

Appreciative Infrastructure for AI-Mediated Organisations: Preventing Algorithmic Closure in Knowledge Management Systems
Authors:

Håkan Sterner
Abstract: Artificial Intelligence (AI) increasingly participates in organisational knowledge work by classifying, summarising, recommending and routing communication. This paper examines how such communication may become organisationally self-reinforcing while losing responsiveness to situated knowing and functionally differentiated grounds. Using an iterative, design-oriented conceptual development approach, the paper develops the Decision-making–Sense-making–Policy-making (DSP) Triad as a three-dimensional architecture. The dimensions concern recursive DSP process coupling, the relation between situated knowing and codified representation, and functional domains of organisational knowing and action. The paper defines algorithmic closure as a limiting trajectory reinforced through selective trace capture, system-reference compression and recursive reuse of enacted selections. As a response, it proposes appreciative infrastructure built around epistemic breakpoints, peer-driven appreciative networks and institutionalised slack. The contribution is to reframe responsible AI in Knowledge Management (KM) as preserving the organisational capacity to reconstruct evidence, contest representations and reopen consequential selections.
Download
Paper Nr: 123
Title:

Analysing the Phenomenon of Refinement and Utilisation of Data, Information and Knowledge
Authors:

Jussi Myllärniemi, Emma Kivimäki, Pasi Hellsten and Essi Halonen
Abstract: The phenomenon of refining and utilising data, information, and knowledge in organisations remains conceptually fragmented and requires further clarification. We studied this phenomenon by conducting a literature review of 15 high-ranking IS and KM journals. Our searches of business analytics, business intelligence, business intelligence and analytics, knowledge-based management, knowledge-based value creation and data-driven decision-making resulted in 1463 articles found from which 74 relevant articles were chosen for the analysis. We analysed the results based on five recognised viewpoints: intelligence, decision-making support, technological systems, process and managerial practice, and management philosophy. The study enabled the identification of key viewpoints of the phenomenon and deepened the understanding of its transversal nature. Our results showed that the seemingly fragmented phenomenon and the concepts related to it share a socio-technical core centered on refining and using data, information, and knowledge for decision-making and value creation. Recent research increasingly emphasizes human, cultural, and organizational capabilities, while intelligence and value creation remain under conceptualized.
Download
Paper Nr: 140
Title:

Sense to Nonsenses: A Propaedeutic Weapon against AI Fault Acceptation
Authors:

Jozef Hvorecký
Abstract: AI easily cheats us. For years, our computers have completed standard calculations without errors. People believe that AI – as it also runs on computers – can hardly make mistakes. It makes them. It takes our input as a basis of its considerations but it “thinks” differently and may come up with the conclusions we would never do – in both positive and negative meaning. When a strange statement comes from a human mouth, our sense to nonsense helps us to unveil its imperfections, to catch the assumed positive meaning or wrong intentions behind it. We do it every day so we are trained. We have to consider the AI outcomes in the same way. AI’s misunderstandings and confusions will always appear. We should learn to face them – and train our pupils to act similarly. The strategy proposed here is based on a strong devotion to nonsenses (faults, misinterpretations, etc.). Intentionally made errors should appear more frequently in all educational sources. The capability of disclosing them will help pupils not only to assess AI outcomes but to improve their interpersonal communication. Several ways of creating additional nonsense and/or absurd statements for educational purposes are presented. Nothing is more vital than error. Bergen Evans.
Download
Paper Nr: 149
Title:

Towards LLM-Powered AI Help Desks for EDIHs: A Service-Orchestration and Knowledge-Access Artefact for Networked Organisations
Authors:

Dmitry Kudryavtsev, Anna Kantosalo, Umair Ali Khan and Akseli Leskinen
Abstract: European Digital Innovation Hubs (EDIHs) help companies and public-sector organisations access digital and AI-related expertise, services and innovation ecosystems. To realise this promising role at scale, EDIHs face two related implementation challenges: a knowledge-management challenge, as customers often lack knowledge of available AI services and EU ecosystem opportunities; and a service-orchestration challenge, because EDIHs are networked organisations comprising multiple stakeholders and must coordinate service selection and delivery across distributed providers, services and infrastructures. AI Help Desks are emerging as first-line support across the EDIH network, but current implementations remain predominantly manual. This position paper reports the first Design Science Research (DSR) iteration towards a generic LLM-powered AI Help Desk artefact for EDIHs. Its practical advancement is to augment manual first-line support with conversational knowledge access and service routing while retaining human escalation. The design contribution lies in combining service-science knowledge about service configuration, orchestration and customer journeys, operationalised through the Data-driven Business-Ecosystem-Skills-Technology (D-BEST) framework, with established LLM mechanisms for retrieval, routing, bounded tool use and response generation. The paper specifies the business use case, translates it into user scenarios and acceptance criteria, and proposes a high-level architecture. The generic artefact is instantiated and demonstrated in the Finnish AI Region (FAIR) EDIH case, where a prototype enables preliminary, primarily formative evaluation.
Download
Paper Nr: 163
Title:

Use of 3D Printed Tools for Better Tactical Overview
Authors:

Jan Drábek, Luděk Rak, František Krompolc, Jaroslav Varecha and Filip Hodulák
Abstract: This paper presents a pilot study examining the impact of three different situational visualization methods on users’ situational awareness, comprehension, and information retention in a military context. The research compares traditional map-based visualization using tactical symbols with two enhanced approaches: maps combined with 3D unit models, and maps combined with both 3D unit models and a 3D terrain model. An exploratory experimental design involving a small sample of military students (n=15) was used to assess recall accuracy related to terrain, friendly forces, enemy forces, and overall situational understanding, supplemented by a qualitative survey evaluating user perception and method preference. Preliminary results indicate that integrating three-dimensional elements, particularly terrain models, tends to improve overall situational awareness and terrain recall, while 3D unit models appear to enhance perception of force disposition but may introduce ambiguity in unit identification. Given the limited sample size and exploratory nature of this study, the findings should be interpreted with caution. Nevertheless, they provide a basis for formulating hypotheses and refining the experimental design for a larger-scale follow-up study. The results suggest that 3D visualization tools have strong potential for training, planning, and educational applications, particularly at higher command levels where preparation time and resources allow their effective deployment.
Download
Paper Nr: 171
Title:

Encoding Unstructured Clinical Free Text to SNOMED CT Using LLM-RAG Architecture
Authors:

Jung X. Lee, Jinie Pak and Yeong-tae Song
Abstract: SNOMED CT is a key clinical terminology used in electronic health record (EHR) systems for interoperability. Use of such standardized clinical coding offers great accuracy and interoperability. However, it is often challenging to encode clinical free-text or doctor notes into SNOMED CT as it is designed for structured fields. So conventional process is done through manual review process requires clinical expert's substantial amount of time, which imposes huge financial burden. Uncoded free text is considered an obstacle for interoperability. Several frameworks have been developed to map clinical text to SNOMED CT concepts; however, many rely primarily on conventional natural language processing techniques developed before the widespread adoption of large language models (LLMs). In this study, we propose a clinical concept normalization architecture that integrates DeepSeek, SapBERT, and FAISS. The framework was evaluated using selected clinical notes from MIMIC-III and biomedical abstracts from MedMentions, with ICD-9 codes and UMLS CUIs serving as gold-standard labels, respectively. Its SNOMED CT predictions were compared with those generated by MedCAT as the baseline system. To enable this comparison across heterogeneous terminology systems, we developed new semantic evaluation metrics (semantic precision and semantic coverage) based on concept mappings and relationships in the UMLS Metathesaurus. In the experiments, our framework achieved higher semantic precision and semantic generation efficiency across both datasets, while MedCAT achieved higher semantic coverage.
Download
Paper Nr: 174
Title:

Semi-Automatic Structural Metadata Generation for Energy Datasets
Authors:

Philipp Schmurr, Andreas Schmidt, Karl-Uwe Stucky, Sebastian Beichter, Wolfgang Suess and Veit Hagenmeyer
Abstract: To satisfy the needs of the energy transition, data-driven solutions as e.g. forecasting or co-simulations, are becoming increasingly important. Energy research can greatly benefit from data publication, as reflected in the requirements of funding agencies. To publish data in a FAIR manner, a core requirement is to additionally publish metadata. However, researchers are often untrained, and no well-established solutions for data publication exist yet. In this paper, we present a software tool that semi-automatically generates structural metadata for an existing dataset. This includes unit annotations and the possibility to interactively add semantic terms from ontologies. The tool is used with three different energy datasets to showcase its application. It is shown that the creation of structural metadata and the annotation with units perform well and greatly reduce the effort to create structural metadata. However, the annotation with semantic terms is still challenging. This is partially based on poor existing terminology and the fact that the semantic annotation is not performed automatically yet.
Download
Paper Nr: 185
Title:

A Value-Added Approach to Integrating Artificial Intelligence in Knowledge Management
Authors:

Kimiz Dalkir
Abstract: This article is a position paper that explores the benefits of integrating artificial intelligence (AI) with core knowledge management (KM) theory and practice. While there are significant challenges associated with AI, more specifically with the new Large Language Models (LLMs) technologies and generative AI (genAI) tools, their potential benefits should not be overlooked. GenAI can add value by increasing inclusivity of otherwise unheard knowledge workers, discovering hidden knowledge legacy content and personalizing knowledge dissemination at a much more nuanced degree. The proposed approach acknowledges the increasingly well-known limitations of genAI tools such as biases, hallucinating and filling in erroneous content to fill knowledge gaps as well as significant privacy concerns. The proposed approach builds on both theoretical and practitioner-focused studies that advocate for contingency plans to mitigate these risks such as human oversight and legislation and recommends a multi-level approach to leverage AI-enabled KM processes that can build trust and validity over time. The strong position taken and defended is that of a hybrid solution where AI adopts the role of an assistant to human decision makers and genAI capabilities add value to KM at the personal, community, organizational and ultimately societal levels.
Download