ahmets

• Avogadro, Roberto; Rauniyar, Ashish; Goknil, Arda; Soylu, Ahmet; Udroiu, Cosmin & Ma, Xiang [Show all 7 contributors for this article] (2026). Human-in-the-Loop Entity Linking Visualization and Refinement with KOALA-UI and the Crocodile Framework. In Chbeir, Richard; Dustdar, Schuhram; Tuan, Tran Ngoc; Manolopoulos, Yannis; Masciari, Elio; Rodriguez, Juan Manual Murillo & Bou-chaaya, Karam (Ed.), Management of Digital EcoSystems. Springer. ISSN 9783032365477. p. 411–424. doi: 10.1007/978-3-032-36547-7_31. Full text in Research Archive Show summary

  Entity linking (EL) on tabular data often produces many ambiguous cell annotations that are costly to review. We present KOALA-UI, a lightweight web interface for inspecting and refining EL results on spreadsheets and CSV-like tables. KOALA-UI provides table navigation, cell-level candidate panels, color-coded scores, simple filtering, and in-place edits with export. To enable reuse across backends, KOALA-UI consumes a compact JSON schema and responses compatible with the W3C Entity Reconciliation specification, allowing integration with potentially any EL backend. We illustrate KOALA-UI with the Crocodile EL framework and demonstrate a human-in-the-loop (HITL) workflow. In addition, we report a preliminary user study, a real business use case, and a validation of pipeline execution through TAO–a workflow engine, where Crocodile was integrated as a module.

• Almasri, Mohammed; Imran, Ali Shariq; Nuci, Krenare; Mavroudi, Anna & Soylu, Ahmet (2026). Exploring Teachers' Perspectives on AI-Assisted Authoring Tool: A TAM-Based Study. IEEE Global Engineering Education Conference, EDUCON. ISSN 2165-9559. doi: 10.1109/EDUCON67543.2026.11574185. Full text in Research Archive
• Almasri, Mohammed; Imran, Ali Shariq; Mavroudi, Anna & Soylu, Ahmet (2026). AI-Powered Framework for developing Virtual labs in Higher Education. IEEE Global Engineering Education Conference, EDUCON. ISSN 2165-9559. doi: 10.1109/EDUCON67543.2026.11574439. Full text in Research Archive
• Ursavaş, Ömer Faruk; Gurcan, Fatih; Durak, Hatice Yıldız; Yalçın, Yasin & Soylu, Ahmet (2026). From E-learning to generative AI: Perspectives on technology acceptance and human behavior in education. Computers in Human Behavior Reports. 22. doi: 10.1016/j.chbr.2026.101117. Full text in Research Archive
• Lyu, Qianhang; Skjæveland, Martin Georg; Li, Siqi; Rao, Yunqing; Soylu, Ahmet & Kiritsis, Dimitris [Show all 8 contributors for this article] (2026). Generating Explainable and Verified Product Information Models From Industrial Documentation. Journal of Computing and Information Science in Engineering. ISSN 1530-9827. 26. doi: 10.1115/1.4071963. Full text in Research Archive
• Klungre, Vidar Norstein; Soylu, Ahmet & Giese, Martin (2026). Avoiding unproductive SPARQL queries through optimized indices. World wide web (Bussum). ISSN 1386-145X. 29(4). doi: 10.1007/s11280-026-01419-6. Full text in Research Archive Show summary

  In interactive query formulation systems for SPARQL, suggested query extensions may result in queries with no answers, so-called unproductive queries. To avoid these, systems must detect and eliminate unproductive extensions. A straightforward approach is to check whether each extended query yields results, but this can be highly inefficient, especially for complex queries involving multiple costly joins. One possible solution is to construct an index that allows efficient checks. However, since queries can be arbitrarily large and complex, it is infeasible to build a complete index within finite space. A practical alternative is to index only the most common query patterns. This produces a smaller, finite index but may introduce inaccuracies in detecting unproductive extensions. The key challenge is balancing precision (the proportion of unproductive extensions correctly identified) against cost (the storage required for the index). In this article, we study this trade-off and present methods to construct indices with low cost while retaining high precision, given a dataset and representative query log. The problem is difficult because the space of possible indices is extremely large, making exhaustive search infeasible. Moreover, computing precision and cost for even a single index can be expensive. We therefore propose four heuristic-based search strategies to explore the index space efficiently, together with approximate cost and precision estimators. We evaluate these methods analytically and experimentally, using a benchmark based on Wikidata. Results show that our approaches identify non-trivial index configurations with high precision and low cost, making them practical for eliminating unproductive extensions in interactive systems.

• Kilinc, Murat; Ursavas, Omer Faruk; Gurcan, Fatih & Soylu, Ahmet (2026). Green and Sustainable Computing: Efficient Strategies, Best Practices and Emerging Research Trends. IEEE Access. doi: 10.1109/ACCESS.2026.3693256. Full text in Research Archive
• Bilal, Mohammad; Ehrenmuller, Katrin; Steindl, Gernot; Zheng, Zhuoxun; Ahmetaj, Shqiponja & Soylu, Ahmet [Show all 8 contributors for this article] (2026). Modeling State Causality in Energy Centred Cyber-Physical-Human Systems With OntoUML. IEEE Access. doi: 10.1109/ACCESS.2026.3683445. Full text in Research Archive
• Palaz, Zeliha Hatipoglu; Cege, Ecem Elif; Maiga, Bamoye; Dalveren, Yaser; Dalveren, Gonca Gokce Menekse & Kara, Ali [Show all 8 contributors for this article] (2026). A Lightweight and Explainable AI Framework Toward Automated Infraocclusion Detection in Pediatric Panoramic Radiographs. Diagnostics (Basel). 16(6). doi: 10.3390/diagnostics16060866. Full text in Research Archive
• Ahmed, Taheera; Ahmed, Ammar; Imran, Ali Shariq; Ullah, Mohib; Manaf, Abdul & Soylu, Ahmet (2026). Exploring the Impact of Self-Supervised Learning and Swin-B Transformer on Saliency Maps and Performance in Multi-Label Chest X-Ray Classification. IEEE Access. 14, p. 21833–21857. doi: 10.1109/ACCESS.2026.3655022. Full text in Research Archive
• Jayawardene, Iroshani; Roman, Dumitru; Ma, Xiang; Liberale, Alessandro; Arce, Alicia & Naranjo, Alfredo Gonzalez [Show all 16 contributors for this article] (2026). Comparative analysis and trends in performance of identical PV panels under different climatic conditions. Solar Energy. ISSN 0038-092X. 307. doi: 10.1016/j.solener.2026.114349. Full text in Research Archive Show summary

  In this paper, a comparative analysis and trends in performance of identical photovoltaic (PV) panels under different climatic conditions are presented. Three PV demo sites situated in Oslo, Norway; Tozeur, Tunisia; and Sevilla, Spain are included into performed comparative analysis. Each site includes sets of identical mono-Si based PV panels supplied by the same manufacturers. The PV panel configurations include: (i) mono-facial, (ii) bi-facial, and (iii) innovative bi-facial PV panels with front glass covered by hydrophobic (HFB) or hydrophilic (HFL) coatings. At all three locations, each panel is equipped with Tigo optimizers, enabling module-level monitoring and DC energy-based performance assessment. Performance metrics are evaluated using IEC 61,724 based indicators, with irradiance derived from global horizontal irradiance data. Statistical analyses are applied to validate observed performance differences between panel types across locations. The results provide a technology-controlled comparison of PV panel behavior across cold-wet, warm-wet, and warm-dry climates, highlighting the influence of temperature, surface reflectivity, and coating type on relative performance trends.

• Gudek, Beyza; Gurcan, Fatih; Soylu, Ahmet & Khan, Akif Quddus (2026). Projecting Türkiye’s CO2 Emissions Future: Multivariate Forecast of Energy–Economy–Environment Interactions and Anthropogenic Drivers. Sustainability. 18(1), p. 471–471. doi: 10.3390/su18010471. Full text in Research Archive Show summary

  Global warming has become a top priority on the international environmental policy agenda. The recent rise in CO2 emissions observed in Türkiye has further emphasized the country’s critical role in addressing climate change. This study aims to estimate Türkiye’s CO2 emissions through 2030 and identify the key socioeconomic and environmental factors driving these emissions, using multiple linear regression (MLR) and time series analysis methods. Six primary variables are examined: population, gross domestic product (GDP), CO2 intensity, per capita energy consumption, total greenhouse gas (GHG) emissions, and forest area. This study introduces a new multivariate forecasting framework that integrates time series projections with multiple linear regression and elasticity-based sensitivity analysis, providing novel insight into the relative influence of key emission drivers compared to prior research. The results suggest that, if current policy trends persist, Türkiye’s CO2 emissions will increase substantially by 2030. Variables such as GHG emissions, energy consumption, and population growth are found to have an increasing effect on emissions, while the limited expansion of forest areas is insufficient to offset this trend. In contrast, the negative correlation between GDP and CO2 emissions suggests that economic growth can occur in alignment with environmental sustainability. The model’s validity is supported by a high R2 (0.99) value and low error rates. The findings indicate that Türkiye must reassess its current strategies and strengthen policies targeting renewable energy, energy efficiency, and carbon sinks to achieve its climate goals. The proposed framework provides a transparent basis for climate planning and policy prioritization in Türkiye.

• Lakhan, Abdullah; Grønli, Tor-Morten; Soylu, Ahmet; Muhammad, Ghulam; Mastoi, Qurat Ul Ain & Wu, Huaming (2025). Transfer Learning-Enabled System for Drone Medicine Delivery Based on Spatio-Temporal Remote Sensing Data in Edge Cloud Networks. IEEE Transactions on Cloud Computing. ISSN 2168-7161. doi: 10.1109/TCC.2025.3639073. Full text in Research Archive
• Gurcan, Fatih & Soylu, Ahmet (2025). Automated waste classification for smart recycling: A multi-class CNN approach with transfer learning and pre-trained models. Environmental Technology & Innovation. 41. doi: 10.1016/j.eti.2025.104673. Full text in Research Archive Show summary

  This study presents a comparative analysis of pre-trained CNN models using transfer learning for the multi-class classification of recyclable waste. The dataset used is publicly available and comprises 15,150 images belonging to 12 different categories of recyclable household waste: paper, cardboard, biological, metal, plastic, green-glass, brown-glass, white-glass, clothes, shoes, batteries, and trash. The methodology involves implementing 14 different CNN models pre-trained on ImageNet (e.g., ConvNeXt_V1, ResNet50, MobileNet_V3_Large) on this dataset through a transfer learning procedure. In all pre-trained models, only the final classifier layer was trained, while all other layers were kept frozen, utilizing a feature extraction strategy that leverages previously learned visual representations. This approach significantly reduced training time and computational cost, while minimizing the risk of overfitting in scenarios with limited data. Model performance was evaluated using classification metrics such as accuracy, F1-score, precision, and recall, as well as computational efficiency indicators including architectural depth, parameter count, and training time. According to the findings, the ConvNeXt_V1 model achieved the highest performance with a test accuracy of 96.95 %. On the other hand, MobileNet_V2 (93.90 %), MobileNet_V3_Large (93.69 %), MNASNet_1_0 (93.69 %), RegNet_Y_800MF (92.87 %), and SqueezeNet_1_0 (92.10 %) offered shorter training times (ranging from 2856 to 7219 s) and stood out as efficient alternatives for embedded system applications.

• Avogadro, Roberto; Makgill, Ian; Thomas, Aleena; Soylu, Ahmet & Roman, Dumitru (2025). Building a Canonical Register of Public Sector Entities: Semantic Linking of Procurement Data at Scale. CEUR Workshop Proceedings. 4085. doi: https:/ceur-ws.org/Vol-4085. Full text in Research Archive Show summary

  Public procurement generates over $13 trillion annually, yet data about public buyers and suppliers remains fragmented, inconsistent, and difficult to link across jurisdictions. This paper presents a practical industrial solution developed by Spend Network within the European project enRichMyData to semantically enrich and reconcile procurement data at scale. The proposed pipeline combines large language models (LLMs) with knowledge graphs (KGs) to create and maintain a canonical register of public sector entities. It supports multilingual, cross-border integration and is designed to serve both public transparency and commercial applications. The pipeline has been evaluated on a manually curated benchmark of 1,000 procurement-related entities and demonstrates high precision and scalability in real-world settings.

• Videsjorden, Adela Nedisan; Nikolov, Nikolay Vladimirov; Lien, Carl-Henrik; Goknil, Arda; Sen, Sagar & Song, Hui [Show all 8 contributors for this article] (2025). Positioning LLM-Enabled Agents as Legal Compliance Aides for Data Pipelines. In Hogan, Aidan; Satoh, Ken; Dağ, Hasan; Turhan, Anni-Yasmin; Roman, Dumitru & Soylu, Ahmet (Ed.), Rules and Reasoning: 9th International Joint Conference, RuleML+RR 2025, Istanbul, Turkey, September 22–24, 2025, Proceedings. Springer. ISSN 9783032088871. p. 227–236. doi: 10.1007/978-3-032-08887-1_14. Full text in Research Archive Show summary

  Ensuring the legal compliance of data pipelines with evolving EU legislation, such as the AI Act, presents a significant challenge due to the complexity of both technical infrastructures and regulatory texts. This paper explores the potential of Large Language Models (LLMs) to support automated compliance assessment of data pipelines, and proposes an approach that leverages LLM-based agents to extract, label, and assess data pipeline artifacts against relevant legal requirements, guided by the actor’s role and the system’s risk level. By decomposing the assessment process into modular agent tasks, we mitigate token limitations and enable fine-grained analysis of regulatory obligations. The approach is supported by a prototype implementation that integrates outputs from SIM-PIPE, a tool for simulating and analyzing big data pipelines, with a structured interpretation of the regulatory document, e.g., the AI Act. The implementation demonstrates the feasibility of intelligent, scalable compliance auditing and highlights key challenges related to trust, context interpretation, and output validity. We argue that such LLM-powered tools can play a critical role in advancing compliance-by-design practices for legally aligned data-driven systems.

• Videsjorden, Adela Nedisan; Song, Hui; Goknil, Arda; Roman, Dumitru & Soylu, Ahmet (2025). LUMEN: Enhancing IoT System Observability with Multi-Agent Large Language Models and Knowledge Graphs. ACM Transactions on Internet of Things (TIOT). ISSN 2691-1914. doi: 10.1145/3772077. Full text in Research Archive Show summary

  The rapid expansion of Internet of Things (IoT) systems has transformed industries through real-time monitoring and automation, generating vast and heterogeneous data streams. As IoT networks expand, the increasing volume and diversity of data, spanning real-time telemetry, device logs, and historical records, complicate the management of IoT systems, including system monitoring, analysis, and reasoning. To address this challenge, we introduce LUMEN (Large Language Models as Unified Multi-Agent Systems for IoT ENhancement), a novel approach combining multi-agent Large Language Models (LLMs), knowledge graphs, and heterogeneous databases to enable cognitive digital twins for IoT observability. LUMEN models IoT systems as knowledge graphs, capturing device relationships and metadata while monitoring data is stored in time-series or object databases. Specialized LLM-based agents collaborate dynamically to analyze IoT systems and explain the findings in natural language, generating and executing analysis code when necessary. Integrated with off-the-shelf network monitoring tools, LUMEN facilitates semantic reasoning and human-in-the-loop collaboration, delivering adaptive insights across diverse data contexts. Two industrial case studies demonstrate the ability of LUMEN to automate analysis workflows, enhance system adaptability, and provide interpretable analytics. This work advances IoT observability by integrating LLMs, semantic intelligence, and explainable analytics into a scalable and adaptive solution using a multi-agent architecture for complex IoT systems.

• Husom, Erik Johannes; Goknil, Arda; Astekin, Merve; Shar, Lwin Khin; Kåsen, Andre & Sen, Sagar [Show all 8 contributors for this article] (2025). Sustainable LLM Inference for Edge AI: Evaluating Quantized LLMs for Energy Efficiency, Output Accuracy, and Inference Latency. ACM Transactions on Internet of Things (TIOT). ISSN 2691-1914. 6(4). doi: 10.1145/3767742. Full text in Research Archive Show summary

  Deploying Large Language Models (LLMs) on edge devices presents significant challenges due to computational constraints, memory limitations, inference speed, and energy consumption. Model quantization has emerged as a key technique to enable efficient LLM inference by reducing model size and computational overhead. In this study, we conduct a comprehensive analysis of 28 quantized LLMs from the Ollama library, which applies by default Post-Training Quantization (PTQ) and weight-only quantization techniques, deployed on an edge device (Raspberry Pi 4 with 4 GB RAM). We evaluate energy efficiency, inference performance, and output accuracy across multiple quantization levels and task types. Models are benchmarked on five standardized datasets (CommonsenseQA, BIG-Bench Hard, TruthfulQA, GSM8K, and HumanEval), and we employ a high-resolution, hardware-based energy measurement tool to capture real-world power consumption. Our findings reveal the trade-offs between energy efficiency, inference speed, and accuracy in different quantization settings, highlighting configurations that optimize LLM deployment for resource-constrained environments. By integrating hardware-level energy profiling with LLM benchmarking, this study provides actionable insights for sustainable AI, bridging a critical gap in existing research on energy-aware LLM deployment.

• Katper, Saif Hassan; Mujtaba, Ghulam; Ullah, Habib; Imran, Ali Shariq; Soylu, Ahmet & Ullah, Mohib (2025). SSF4VSU: A Self-Supervised Synergetic Framework for Visual Scene Understanding. IEEE Access. 13, p. 197544–197561. doi: 10.1109/access.2025.3634778. Full text in Research Archive Show summary

  Visual scene understanding involves multiple tasks, including single-object tracking (SOT), multi-object tracking (MOT), video object segmentation (VOS), and its Multi-Object Tracking and Segmentation (MOTS) variant, which have traditionally been studied in isolation. Such fragmentation leads to task-specific architectures that must be retrained for each new scenario and rely heavily on annotated data. This work proposes a synergetic model called SSF4VSU that simultaneously addresses SOT, MOT, VOS, and MOTS. SSF4VSU employs a shared backbone with a unified embedding space for different tasks, a Temporal Attention Module (TAM) to align features across frames and resist occlusions, a Temporal Consistency Module (TCM) to enforce smooth identity preservation, and a self-supervised learning (SSL) branch that leverages unlabeled video sequences for improved generalization. Training follows a multi-task curriculum with dynamic loss balancing, demonstrating efficiency and scalability. A comprehensive evaluation across six public benchmarks shows that SSF4VSU surpasses specialised and unified state-of-the-art models. On SOT benchmarks it achieves 74.7% success AUC and 80.4% precision on LaSOT, and 85.8% AUC and 84.3% precision on TrackingNet, matching or exceeding recent specialised trackers. For MOT, it records 82.9% MOTA and 83.3% IDF1 on the MOT17 dataset, while on BDD100K it generalises to driving scenes without task-specific tuning. On VOS benchmarks, SSF4VSU delivers 93.3% J&F on DAVIS-2016 and 89.0% J&F on DAVIS-2017, surpassing strong segmentation methods and unified baselines. In the MOTS setting it achieves 69.0% sMOTSA on MOTS20 and 31.2% mMOTSA on BDD100K MOTS. Ablation studies reveal that TAM is critical for accurate localisation, TCM is indispensable for identity stability, and SSL consistently improves generalization. The results demonstrate that a single unified model can perform diverse tracking and segmentation tasks without sacrificing accuracy or efficiency. By reducing reliance on labelled data, improving temporal reasoning, and ensuring interpretability, SSF4VSU advances the state of visual scene understanding and lays the foundation for general-purpose, multi-task video analysis.

• Kilinc, Murat; Gurcan, Fatih & Soylu, Ahmet (2025). LLM-Based Generative AI in Medicine: Analysis of Current Research Trends With BERTopic. IEEE Access. 13, p. 157026–157043. doi: 10.1109/access.2025.3606335. Full text in Research Archive Show summary

  Recent advances in generative artificial intelligence (GenAI) have made large language models (LLMs) transformative tools in the healthcare industry. Powered by deep learning, these models show great potential in many medical applications such as clinical decision support systems, biomedical text mining and personalized patient care. However, while research on the use of LLMs in medicine is growing rapidly, there is a need to systematically analyze the key themes, emerging trends and challenges in this field. This study aims to identify dominant themes, track research trends and identify gaps in the literature by utilizing a large dataset of scientific publications in the medical field. A total of 3,941 academic publications related to the use of LLMs in medicine were retrieved from the Scopus database, covering the period from 2023 to 2024. Accordingly, the BERTopic method, an advanced topic modeling technique, is used to analyze and classify publications on LLMs applications in medicine. The findings show that LLMs are mostly concentrated in the fields of radiology, ophthalmology and mental health and that these models have the potential to support clinical processes. In particular, LLMs have been found to have a great impact on the analysis of medical imaging reports, are used for early diagnosis of ophthalmologic diseases, and are prominent in depression and suicide risk assessments in mental health. This study provides valuable insights for healthcare professionals, GenAI researchers, and policy makers, laying a solid foundation for future research and strategic applications for the use of LLMs in medicine. While emphasizing the transformative impact of LLMs in healthcare, it also highlights the necessity of responsible GenAI applications.

• Novalija, Inna; Roman, Dumitru; Belotti, Federico; Alexiev, Vladimir; Rei, Luis & Avogadro, Roberto [Show all 14 contributors for this article] (2025). From Code to Concept: A Semantic Approach to AI Innovation Discovery in Open Source Software Repositories. IEEE Access. 13, p. 129995–130014. doi: 10.1109/access.2025.3590135. Full text in Research Archive Show summary

  Artificial Intelligence (AI) is a transformative force driving innovation, yet tracking AI-related advancements remains challenging due to the rapid pace of development and unstructured data from platforms like GitHub. This paper proposes an AI-driven approach to innovation detection, leveraging GitHub as a data source to systematically identify and link AI projects to organizations. Key contributions include a domain-specific taxonomy comprising 7,490 AI topics, a modular pipeline for semantic annotation and entity linking, and a trend detection framework based on Singular Spectrum Analysis (SSA). A knowledge graph is constructed to represent relationships among AI topics, projects, and companies, thereby enabling structured innovation tracking. The approach addresses challenges such as data sparsity and noise, demonstrating strengths in semantic annotation and topic categorization. Results highlight the potential for accurately detecting AI innovations and linking them to organizational entities, offering valuable insights for researchers, companies, and policymakers. This work contributes a scalable, automated approach for AI innovation tracking, with future directions focusing on refining entity linking and expanding the knowledge graph to capture emerging trends.

• Teigen, Håvard; Ahmed, Ammar; Imran, Ali Shariq; Ullah, Mohib; Azad, R. Muhammad Atif & Soylu, Ahmet (2025). From Blurs to Birds: Localization and Classification of Hard-to-See Bird Species in Norwegian Wilderness Camera Trap Images. IEEE Access. 13, p. 169534–169553. doi: 10.1109/ACCESS.2025.3613068. Full text in Research Archive
• Gurcan, Fatih; Soylu, Ahmet & Khan, Akif Quddus (2025). Towards a Sustainable Workforce in Big Data Analytics: Skill Requirements Analysis from Online Job Postings Using Neural Topic Modeling. Sustainability. 17(20), p. 9293–9293. doi: 10.3390/su17209293. Full text in Research Archive Show summary

  Big data analytics has become a cornerstone of modern industries, driving advancements in business intelligence, competitive intelligence, and data-driven decision-making. This study applies Neural Topic Modeling (NTM) using the BERTopic framework and N-gram-based textual content analysis to examine job postings related to big data analytics in real-world contexts. A structured analytical process was conducted to derive meaningful insights into workforce trends and skill demands in the big data analytics domain. First, expertise roles and tasks were identified by analyzing job titles and responsibilities. Next, key competencies were categorized into analytical, technical, developer, and soft skills and mapped to corresponding roles. Workforce characteristics such as job types, education levels, and experience requirements were examined to understand hiring patterns. In addition, essential tasks, tools, and frameworks in big data analytics were identified, providing insights into critical technical proficiencies. The findings show that big data analytics requires expertise in data engineering, machine learning, cloud computing, and AI-driven automation. They also emphasize the importance of continuous learning and skill development to sustain a future-ready workforce. By connecting academia and industry, this study provides valuable implications for educators, policymakers, and corporate leaders seeking to strengthen workforce sustainability in the era of big data analytics.

• Jayawardene, Iroshani; Skrzycki, Magdalena; Ma, Xiang; Soylu, Ahmet & Roman, Dumitru (2025). Synthetic Data Generation for Electric Vehicle Charging, Workshop on AI for Sustainable Energy Systems and Green AI (Book of Abstracts). Verlag der Technischen Universität Graz. ISSN 9783991610458. p. 21–22. doi: 10.3217/978-3-99161-045-8. Full text in Research Archive Show summary

  Konferanseabstrakt.

• Disci, Rukiye; Gurcan, Fatih & Soylu, Ahmet (2025). Advanced Brain Tumor Classification in MR Images Using Transfer Learning and Pre-Trained Deep CNN Models. Cancers. 17(1). doi: 10.3390/cancers17010121. Full text in Research Archive
• Jayawardene, Iroshani; Avogadro, Roberto; Soylu, Ahmet & Roman, Dumitru (2024). TabLinkLLM: An LLM-Based Approach for Entity Linking in Tabular Data. In Chan, Jonathan; Tan, Ah-Hwee & Yao, Yiyu (Ed.), 2024 IEEE/WIC International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798331504946. p. 206–214. doi: 10.1109/wi-iat62293.2024.00035. Full text in Research Archive Show summary

  Entity linking (EL) involves identifying references to entities in source data, which can be structured or unstructured, and associating them with their respective records in a structured knowledge base. This technique is beneficial when applied to tabular data, facilitating tasks such as data integration, business intelligence, and the construction of knowledge graphs. Current EL algorithms for tabular data often face challenges such as ambiguity, heterogeneity, and limited context. In this paper, we propose TabLinkLLM - a generic approach for tabular data entity linking using Large Language Models (LLMs), specifically optimized through prompt engineering, retrieval-augmented generation, and fine-tuning. Our experimental comparisons with leading EL models reveal that while our approach does not outperform specialized EL models in terms of performance, the broad knowledge base of LLMs proves advantageous in addressing scenarios that these specialized models cannot handle.

• Khan, Akif Quddus; Matskin, Mihhail; Prodan, Radu; Bussler, Christoph; Roman, Dumitru & Soylu, Ahmet (2024). Cloud storage cost: a taxonomy and survey. World wide web (Bussum). ISSN 1386-145X. 27(4). doi: 10.1007/s11280-024-01273-4. Full text in Research Archive Show summary

  Cloud service providers offer application providers with virtually infinite storage and computing resources, while providing cost-efficiency and various other quality of service (QoS) properties through a storage-as-a-service (StaaS) approach. Organizations also use multicloud or hybrid solutions by combining multiple public and/or private cloud service providers to avoid vendor lock-in, achieve high availability and performance, and optimise cost. Indeed cost is one of the important factors for organizations while adopting cloud storage; however, cloud storage providers offer complex pricing policies, including the actual storage cost and the cost related to additional services (e.g., network usage cost). In this article, we provide a detailed taxonomy of cloud storage cost and a taxonomy of other QoS elements, such as network performance, availability, and reliability. We also discuss various cost trade-offs, including storage and computation, storage and cache, and storage and network. Finally, we provide a cost comparison across different storage providers under different contexts and a set of user scenarios to demonstrate the complexity of cost structure and discuss existing literature for cloud storage selection and cost optimization. We aim that the work presented in this article will provide decision-makers and researchers focusing on cloud storage selection for data placement, cost modelling, and cost optimization with a better understanding and insights regarding the elements contributing to the storage cost and this complex problem domain.

• Gurcan, Fatih & Soylu, Ahmet (2024). Synthetic Boosted Resampling Using Deep Generative Adversarial Networks: A Novel Approach to Improve Cancer Prediction from Imbalanced Datasets. Cancers. 16(23). doi: 10.3390/cancers16234046. Full text in Research Archive
• Khan, Akif Quddus; Nikolov, Nikolay Vladimirov; Matskin, Mihhail; Prodan, Radu; Bussler, Christoph & Roman, Dumitru [Show all 7 contributors for this article] (2024). A Taxonomy for Cloud Storage Cost. Communications in Computer and Information Science (CCIS). ISSN 1865-0929. 2022, p. 317–330. doi: 10.1007/978-3-031-51643-6_23. Full text in Research Archive Show summary

  The cost of using cloud storage services is complex and often an unclear structure, while it is one of the important factors for organisations adopting cloud storage. Furthermore, organisations take advantage of multi-cloud or hybrid solutions to combine multiple public and/or private cloud service providers to avoid vendor lock-in, achieve high availability and performance, optimise cost, etc. This complicated ecosystem makes it even harder to understand and manage cost. Therefore, in this paper, we provide a taxonomy of cloud storage cost in order to provide a better understanding and insights on this complex problem domain.

• Gurcan, Fatih & Soylu, Ahmet (2024). Learning from Imbalanced Data: Integration of Advanced Resampling Techniques and Machine Learning Models for Enhanced Cancer Diagnosis and Prognosis. Cancers. 16(19). doi: 10.3390/cancers16193417. Full text in Research Archive
• Zheng, Zhuoxun; Zhou, Baifan; Yang, Hui; Tan, Zhipeng; Waaler, Arild Torolv Søetorp & Kharlamov, Evgeny [Show all 7 contributors for this article] (2024). Low-Dimensional Hyperbolic Knowledge Graph Embedding for Better Extrapolation to Under-Represented Data. Lecture Notes in Computer Science (LNCS). ISSN 0302-9743. Full text in Research Archive
• Khan, Akif Quddus; Matskin, Mihhail; Prodan, Radu; Bussler, Christoph; Roman, Dumitru & Soylu, Ahmet (2024). Cost modelling and optimisation for cloud: a graph-based approach. Journal of Cloud Computing. 13(1). doi: 10.1186/s13677-024-00709-6. Full text in Research Archive Show summary

  Cloud computing has become popular among individuals and enterprises due to its convenience, scalability, and flexibility. However, a major concern for many cloud service users is the rising cost of cloud resources. Since cloud computing uses a pay-per-use model, costs can add up quickly, and unexpected expenses can arise from a lack of visibility and control. The cost structure gets even more complicated when working with multi-cloud or hybrid environments. Businesses may spend much of their IT budget on cloud computing, and any savings can improve their competitiveness and financial stability. Hence, an efficient cloud cost management is crucial. To overcome this difficulty, new approaches and tools are being developed to provide greater oversight and command over cloud a graph-based approach for modelling cost elements and cloud resources and a potential way to solve the resulting constraint problem of cost optimisation. In this context, we primarily consider utilisation, cost, performance, and availability. The proposed approach is evaluated on three different user scenarios, and results indicate that it could be effective in cost modelling, cost optimisation, and scalability. This approach will eventually help organisations make informed decisions about cloud resource placement and manage the costs of software applications and data workflows deployed in single, hybrid, or multi-cloud environments.

• Chen, Qiaosheng; Luo, Weiqing; Huang, Zixian; Lin, Tengteng; Wang, Xiaxia & Soylu, Ahmet [Show all 10 contributors for this article] (2024). ACORDAR 2.0: A Test Collection for Ad Hoc Dataset Retrieval with Densely Pooled Datasets and Question-Style Queries. In Yang, Grace Hui; Wang, Hongning; Han, Sam; Hauff, Claudia; Zuccon, Guido & Zhang, Yi (Ed.), SIGIR '24 : Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval. Association for Computing Machinery (ACM). ISSN 9798400704314. p. 303–312. Full text in Research Archive
• Khan, Akif Quddus; Matskin, Mihhail; Prodan, Radu; Bussler, Christoph; Roman, Dumitru & Soylu, Ahmet (2024). Cloud storage tier optimization through storage object classification. Computing. ISSN 0010-485X. 106, p. 3389–3418. doi: 10.1007/s00607-024-01281-2. Full text in Research Archive Show summary

  Cloud storage adoption has increased over the years given the high demand for fast processing, low access latency, and ever-increasing amount of data being generated by, e.g., Internet of Things applications. In order to meet the users’ demands and provide a cost-effective solution, cloud service providers offer tiered storage; however, keeping the data in one tier is not cost-effective. In this respect, cloud storage tier optimization involves aligning data storage needs with the most suitable and cost-effective storage tier, thus reducing costs while ensuring data availability and meeting performance requirements. Ideally, this process considers the trade-off between performance and cost, as different storage tiers offer different levels of performance and durability. It also encompasses data lifecycle management, where data is automatically moved between tiers based on access patterns, which in turn impacts the storage cost. In this respect, this article explores two novel classification approaches, rule-based and game theory-based, to optimize cloud storage cost by reassigning data between different storage tiers. Four distinct storage tiers are considered: premium, hot, cold, and archive. The viability and potential of the proposed approaches are demonstrated by comparing cost savings and analyzing the computational cost using both fully-synthetic and semi-synthetic datasets with static and dynamic access patterns. The results indicate that the proposed approaches have the potential to significantly reduce cloud storage cost, while being computationally feasible for practical applications. Both approaches are lightweight and industry- and platform-independent.

• Zheng, Zhuoxun; Zhou, Baifan; Yang, Hui; Tan, Zhipeng; Sun, Zequn & Li, Chunnong [Show all 9 contributors for this article] (2024). Knowledge graph embedding closed under composition. Data mining and knowledge discovery. ISSN 1384-5810. 38. doi: 10.1007/s10618-024-01050-x. Full text in Research Archive Show summary

  Knowledge Graph Embedding (KGE) has attracted increasing attention. Relation patterns, such as symmetry and inversion, have received considerable focus. Among them, composition patterns are particularly important, as they involve nearly all relations in KGs. However, prior KGE approaches often consider relations to be compositional only if they are well-represented in the training data. Consequently, it can lead to performance degradation, especially for under-represented composition patterns. To this end, we propose HolmE, a general form of KGE with its relation embedding space closed under composition, namely that the composition of any two given relation embeddings remains within the embedding space. This property ensures that every relation embedding can compose, or be composed by other relation embeddings. It enhances HolmE’s capability to model under-represented (also called long-tail) composition patterns with limited learning instances. To our best knowledge, our work is pioneering in discussing KGE with this property of being closed under composition. We provide detailed theoretical proof and extensive experiments to demonstrate the notable advantages of HolmE in modelling composition patterns, particularly for long-tail patterns. Our results also highlight HolmE’s effectiveness in extrapolating to unseen relations through composition and its state-of-the-art performance on benchmark datasets.

• Avogadro, Roberto; Jayawardene, Iroshani; Ma, Xiang; Soylu, Ahmet & Roman, Dumitru (2024). Koala-UI: An Interactive User Interface for Tabular Data Linking. CEUR Workshop Proceedings. 3816. doi: https:/ceur-ws.org/Vol-3816/. Full text in Research Archive Show summary

  This paper introduces Koala-UI – a user interface system aimed at simplifying the entity linking process within data enrichment pipelines. Koala-UI provides an intuitive mechanism for linking entities across datasets, combining automation with human feedback to ensure accurate and consistent data. Koala-UI was successfully applied in use cases such as public procurement, where it enabled enrichment of a tenders dataset by linking entities to external knowledge graphs. Future developments will focus on expanding its backend to support additional models and enhance its human-in-the-loop capabilities.

• Jayawardene, Iroshani; Roman, Dumitru; Zhao, Yingxiang; Ulyashin, Alexsander G.; Soylu, Ahmet & Ma, Xiang (2024). Towards an Open Energy Management System for Integrated Energy Storage and Electric Vehicle Fast Charging Station. CEUR Workshop Proceedings. 3816, p. 1–7. doi: https:/ceur-ws.org/Vol-3816/. Full text in Research Archive Show summary

  SINTEF’s solar energy infrastructure, which integrates photovoltaic (PV) panels, meteorological instruments, inverters, and data pipelines, enables real-time data acquisition and visualization. Building on this infrastructure, we developed AI-based algorithms for forecasting PV power output and analyzing the performance of the system and various PV panel types. The potential of this system is significantly enhanced by integrating battery storage, electric vehicle (EV) charging stations, external energy market pricing, advanced meteorological forecasting, and demand forecasting. These integrations support the development of a comprehensive open energy management system (EMS) that facilitates localized energy production and adaptive demand response. In this paper, we outline the key elements of an open EMS that includes PV, batteries, and EV charging station. We describe a prototype and discuss further developments needed in this area. Our approach leverages machine learning to optimize energy flow decisions, with the inclusion of rule-based models to guide and explain these decisions. This work addresses the gap in applying theoretical open EMS models to practical residential and commercial settings, aiming to provide a dynamic platform where forecasting and optimization methods are refined and implemented in real-world scenarios.

• Khan, Akif Quddus; Nikolov, Nikolay Vladimirov; Matskin, Mihhail; Prodan, Radu; Bussler, Christoph & Roman, Dumitru [Show all 7 contributors for this article] (2023). Towards Graph-based Cloud Cost Modelling and Optimisation. In Cuzzocrea, Alfredo; Sharmin, Moushumi; Teranishi, Yuuichi & Towey, Dave (Ed.), 2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798350326970. doi: 10.1109/COMPSAC57700.2023.00203. Full text in Research Archive Show summary

  Cloud computing has become an increasingly popular choice for businesses and individuals due to its flexibility, scalability, and convenience; however, the rising cost of cloud resources has become a significant concern for many. The payper-use model used in cloud computing means that costs can accumulate quickly, and the lack of visibility and control can result in unexpected expenses. The cost structure becomes even more complicated when dealing with hybrid or multi-cloud environments. For businesses, the cost of cloud computing can be a significant portion of their IT budget, and any savings can lead to better financial stability and competitiveness. In this respect, it is essential to manage cloud costs effectively. This requires a deep understanding of current resource utilization, forecasting future needs, and optimising resource utilization to control costs. To address this challenge, new tools and techniques are being developed to provide more visibility and control over cloud computing costs. In this respect, this paper explores a graphbased solution for modelling cost elements and cloud resources and potential ways to solve the resulting constraint problem of cost optimisation. We primarily consider utilization, cost, performance, and availability in this context. Such an approach will eventually help organizations make informed decisions about cloud resource placement and manage the costs of software applications and data workflows deployed in single, hybrid, or multi-cloud environments.

• Layegh, Amirhossein; Payberah, Amir Hossein; Soylu, Ahmet; Roman, Dumitru & Matskin, Mihhail (2023). ContrastNER: Contrastive-based Prompt Tuning for Few-shot NER. In Cuzzocrea, Alfredo; Sharmin, Moushumi; Teranishi, Yuuichi & Towey, Dave (Ed.), 2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798350326970. doi: 10.1109/COMPSAC57700.2023.00038. Full text in Research Archive Show summary

  Prompt-based language models have produced encouraging results in numerous applications, including Named Entity Recognition (NER) tasks. NER aims to identify entities in a sentence and provide their types. However, the strong performance of most available NER approaches is heavily dependent on the design of discrete prompts and a verbalizer to map the model-predicted outputs to entity categories, which are complicated undertakings. To address these challenges, we present ContrastNER, a prompt-based NER framework that employs both discrete and continuous tokens in prompts and uses a contrastive learning approach to learn the continuous prompts and forecast entity types. The experimental results demonstrate that ContrastNER obtains competitive performance to the state-of-the-art NER methods in high-resource settings and outperforms the state-of-the-art models in low-resource circumstances without requiring extensive manual prompt engineering and verbalizer design.

• Tan, Zhipeng; Zhou, Baifan; Zheng, Zhuoxun; Savkovic, Ognjen; Huang, Ziqiang & Gonzalez, Irlan-Grangel [Show all 8 contributors for this article] (2023). Literal-Aware Knowledge Graph Embedding for Welding Quality Monitoring: A Bosch Case. Lecture Notes in Computer Science (LNCS). ISSN 0302-9743. 14266, p. 453–471. doi: 10.1007/978-3-031-47243-5_25. Full text in Research Archive Show summary

  Recently there has been a series of studies in knowledge graph embedding (KGE), which attempts to learn the embeddings of the entities and relations as numerical vectors and mathematical mappings via machine learning (ML). However, there has been limited research that applies KGE for industrial problems in manufacturing. This paper investigates whether and to what extent KGE can be used for an important problem: quality monitoring for welding in manufacturing industry, which is an impactful process accounting for production of millions of cars annually. The work is in line with Bosch research of data-driven solutions that intends to replace the traditional way of destroying cars, which is extremely costly and produces waste. The paper tackles two very challenging questions simultaneously: how large the welding spot diameter is; and to which car body the welded spot belongs to. The problem setting is difficult for traditional ML because there exist a high number of car bodies that should be assigned as class labels. We formulate the problem as link prediction, and experimented popular KGE methods on real industry data, with consideration of literals. Our results reveal both limitations and promising aspects of adapted KGE methods.

• Zhou, Baifan; Nikolov, Nikolay Vladimirov; Zheng, Zhuoxun; Luo, Xianghui; Savkovic, Ognjen & Roman, Dumitru [Show all 8 contributors for this article] (2023). Scaling Data Science Solutions with Semantics and Machine Learning: Bosch Case. Lecture Notes in Computer Science (LNCS). ISSN 0302-9743. 14266, p. 380–399. doi: 10.1007/978-3-031-47243-5_21. Full text in Research Archive Show summary

  Industry 4.0 and Internet of Things (IoT) technologies unlock unprecedented amount of data from factory production, posing big data challenges in volume and variety. In that context, distributed computing solutions such as cloud systems are leveraged to parallelise the data processing and reduce computation time. As the cloud systems become increasingly popular, there is increased demand that more users that were originally not cloud experts (such as data scientists, domain experts) deploy their solutions on the cloud systems. However, it is non-trivial to address both the high demand for cloud system users and the excessive time required to train them. To this end, we propose SemCloud, a semantics-enhanced cloud system, that couples cloud system with semantic technologies and machine learning. SemCloud relies on domain ontologies and mappings for data integration, and parallelises the semantic data integration and data analysis on distributed computing nodes. Furthermore, SemCloud adopts adaptive Datalog rules and machine learning for automated resource configuration, allowing non-cloud experts to use the cloud system. The system has been evaluated in industrial use case with millions of data, thousands of repeated runs, and domain users, showing promising results.

• Benvenuti, Dario; Marrella, Andrea; Rossi, Jacopo; Nikolov, Nikolay Vladimirov; Roman, Dumitru & Soylu, Ahmet [Show all 7 contributors for this article] (2023). A Reference Data Model to Specify Event Logs for Big Data Pipeline Discovery. Lecture Notes in Business Information Processing. ISSN 1865-1348. 490, p. 38–54. doi: 10.1007/978-3-031-41623-1_3. Full text in Research Archive Show summary

  State-of-the-art approaches for managing Big Data pipelines assume their anatomy is known by design and expressed through adhoc Domain-Specific Languages (DSLs), with insufficient knowledge of the dark data involved in the pipeline execution. Dark data is data that organizations acquire during regular business activities but is not used to derive insights or for decision-making. The recent literature on Big Data processing agrees that a new breed of Big Data pipeline discovery (BDPD) solutions can mitigate this issue by solely analyzing the event log that keeps track of pipeline executions over time. Relying on well-established process mining techniques, BDPD can reveal fact-based insights into how data pipelines transpire and access dark data. However, to date, a standard format to specify the concept of Big Data pipeline execution in an event log does not exist, making it challenging to apply process mining to achieve the BDPD task. To address this issue, in this paper we formalize a universally applicable reference data model to conceptualize the core properties and attributes of a data pipeline execution. We provide an implementation of the model as an extension to the XES interchange standard for event logs, demonstrate its practical applicability in a use case involving a data pipeline for managing digital marketing campaigns, and evaluate its effectiveness in uncovering dark data manipulated during several pipeline executions.

• Khan, Akif Quddus; Nikolov, Nikolay Vladimirov; Matskin, Mihhail; Prodan, Radu; Bussler, Christoph & Roman, Dumitru [Show all 7 contributors for this article] (2023). Towards Cloud Storage Tier Optimization with Rule-Based Classification. Lecture Notes in Computer Science (LNCS). ISSN 0302-9743. 14183, p. 205–216. doi: 10.1007/978-3-031-46235-1_13. Full text in Research Archive Show summary

  Cloud storage adoption has increased over the years as more and more data has been produced with particularly high demand for fast processing and low latency. To meet the users’ demands and to provide a cost-effective solution, cloud service providers (CSPs) have offered tiered storage; however, keeping the data in one tier is not a cost-effective approach. Hence, several two-tiered approaches have been developed to classify storage objects into the most suitable tier. In this respect, this paper explores a rule-based classification approach to optimize cloud storage cost by migrating data between different storage tiers. Instead of two, four distinct storage tiers are considered, including premium, hot, cold, and archive. The viability and potential of the approach are demonstrated by comparing cost savings achieved when data was moved between tiers versus when it remained static. The results indicate that the proposed approach has the potential to significantly reduce cloud storage cost, thereby providing valuable insights for organizations seeking to optimize their cloud storage strategies. Finally, the limitations of the proposed approach are discussed along with the potential directions for future work, particularly the use of game theory to incorporate a feedback loop to extend and improve the proposed approach accordingly.

• Tan, Zhipeng; Zheng, Zhuoxun; Klironomos, Antonis; Gad-Elrab, Mohamed H.; Xiao, Guohui & Soylu, Ahmet [Show all 8 contributors for this article] (2023). Literal-Aware Knowledge Graph Embedding for Welding Quality Monitoring. CEUR Workshop Proceedings. 3632. Full text in Research Archive Show summary

  Recently there has been a series of studies in knowledge graph embedding (KGE), which attempts to learn the embeddings of the entities and relations as numerical vectors and mathematical mappings via machine learning (ML). However, there has been limited research that applies KGE for industrial problems in manufacturing. This paper investigates whether and to what extent KGE can be used for an important problem, that is quality monitoring for welding in manufacturing industry. It is an important process accounting for production of millions of cars annually. The work is in line with our research of data-driven solutions that intends to replace the traditional costly quality monitoring. The paper tackles two challenging questions simultaneously: how large the welding spot diameter is; and to which car body the welded spot belongs to. The problem setting is difficult for traditional ML because there exist a high number of car bodies that should be assigned as class labels. We formulate the problem as link prediction, and experimented popular KGE methods with literals on real industry data, with consideration of literals. This paper accompanies the full paper in in-use track and provides additional discussion on problem formulation, literal handling strategies, and included information in industrial KG construction.

• Khan, Akif Quddus; Nikolov, Nikolay; Matskin, Minhail; Prodan, Radu; Roman, Dumitru & Sahin, Bekir [Show all 8 contributors for this article] (2023). Smart Data Placement Using Storage-as-a-Service Model for Big Data Pipelines. Sensors. 23(2). doi: 10.3390/s23020564. Full text in Research Archive Show summary

  Big data pipelines are developed to process data characterized by one or more of the three big data features, commonly known as the three Vs (volume, velocity, and variety), through a series of steps (e.g., extract, transform, and move), making the ground work for the use of advanced analytics and ML/AI techniques. Computing continuum (i.e., cloud/fog/edge) allows access to virtually infinite amount of resources, where data pipelines could be executed at scale; however, the implementation of data pipelines on the continuum is a complex task that needs to take computing resources, data transmission channels, triggers, data transfer methods, integration of message queues, etc., into account. The task becomes even more challenging when data storage is considered as part of the data pipelines. Local storage is expensive, hard to maintain, and comes with several challenges (e.g., data availability, data security, and backup). The use of cloud storage, i.e., storage-as-a-service (StaaS), instead of local storage has the potential of providing more flexibility in terms of scalability, fault tolerance, and availability. In this article, we propose a generic approach to integrate StaaS with data pipelines, i.e., computation on an on-premise server or on a specific cloud, but integration with StaaS, and develop a ranking method for available storage options based on five key parameters: cost, proximity, network performance, server-side encryption, and user weights/preferences. The evaluation carried out demonstrates the effectiveness of the proposed approach in terms of data transfer performance, utility of the individual parameters, and feasibility of dynamic selection of a storage option based on four primary user scenarios.

• Zheng, Zhuoxun; Savkovic, Ognjen; Nikolov, Nikolay Vladimirov; Luu, Huu Phuc; Soylu, Ahmet & Kharlamov, Evgeny [Show all 7 contributors for this article] (2023). Datalog with External Machine Learning Functions for Automated Cloud Resource Configuration. CEUR Workshop Proceedings. 3632. doi: https:/ceur-ws.org/Vol-3632/ISWC2023_paper_480.pdf. Full text in Research Archive Show summary

  Industry 4.0 and Internet of Things (IoT) technologies unlock unprecedented amount of data from factory production, posing big data challenges. In that context, distributed computing solutions such as cloud systems are leveraged to parallelise the data processing and reduce computation time. As the cloud systems become increasingly popular, there is increased demand that more users that were originally not cloud experts (such as data scientists, domain experts) deploy their solutions on the cloud systems. To this end, we propose SemCloud, a semantics-enhanced cloud system, for tackling the challenges of data volume and more users. The system has been evaluated in industrial use case with millions of data, thousands of repeated runs, and domain users, showing promising results. This poster paper accompanies our full paper and focuses on Datalog rules with external machine learning functions for automated resource configuration, and provides additional discussion on formalism and implementation techniques.

• Zheng, Zhuoxun; Zhou, Baifan; Tan, Zhipeng; Savkovic, Ognjen; Rincon-Yanez, Diego & Nikolov, Nikolay Vladimirov [Show all 9 contributors for this article] (2023). Semantic Cloud System for Scaling Data Science Solutions for Welding at Bosch. CEUR Workshop Proceedings. 3632. doi: https:/ceur-ws.org/Vol-3632/. Full text in Research Archive Show summary

  Background and Challenges. Industry 4.0 focuses on smart factories that rely on IoT tech- nology for automation. This produces massive amounts of production data, increasing the demand for data-driven solutions and cloud technology. Yet, users of these solutions and cloud technology are often not cloud experts, such as domain experts and data scientists (Fig. 1.1). In a standard setting of a data science project, the team requires extensive assistance from cloud experts, whenever they want to deploy solutions or make small changes to their solutions deployed on the cloud. To facilitate the adoption of cloud systems careful planing to balance cost and benefits is required. Scaling data science solutions presents challenges of handling high data volume and enabling a broader users which are non-cloud experts to use cloud systems.

• Yilmaz, Vadi Su; Akdag, Metehan; Dalveren, Yaser; Doruk, Resat Ozgur; Kara, Ali & Soylu, Ahmet (2023). Investigating the Impact of Two Major Programming Environments on the Accuracy of Deep Learning-Based Glioma Detection from MRI Images. Diagnostics (Basel). 13(4). doi: 10.3390/diagnostics13040651. Full text in Research Archive Show summary

  Brain tumors have been the subject of research for many years. Brain tumors are typically classified into two main groups: benign and malignant tumors. The most common tumor type among malignant brain tumors is known as glioma. In the diagnosis of glioma, different imaging technologies could be used. Among these techniques, MRI is the most preferred imaging technology due to its high-resolution image data. However, the detection of gliomas from a huge set of MRI data could be challenging for the practitioners. In order to solve this concern, many Deep Learning (DL) models based on Convolutional Neural Networks (CNNs) have been proposed to be used in detecting glioma. However, understanding which CNN architecture would work efficiently under various conditions including development environment or programming aspects as well as performance analysis has not been studied so far. In this research work, therefore, the purpose is to investigate the impact of two major programming environments (namely, MATLAB and Python) on the accuracy of CNN-based glioma detection from Magnetic Resonance Imaging (MRI) images. To this end, experiments on the Brain Tumor Segmentation (BraTS) dataset (2016 and 2017) consisting of multiparametric magnetic MRI images are performed by implementing two popular CNN architectures, the three-dimensional (3D) U-Net and the V-Net in the programming environments. From the results, it is concluded that the use of Python with Google Colaboratory (Colab) might be highly useful in the implementation of CNN-based models for glioma detection. Moreover, the 3D U-Net model is found to perform better, attaining a high accuracy on the dataset. The authors believe that the results achieved from this study would provide useful information to the research community in their appropriate implementation of DL approaches for brain tumor detection.

• Tahmasebi, Shirin; Payberah, Amir H.; Soylu, Ahmet; Roman, Dumitru & Matskin, Mihhail (2023). TRANSQLATION: TRANsformer-based SQL RecommendATION. In He, Jingrui; Palpanas, Themis; Hu, Xiaohua; Cuzzocrea, Alfredo; Dou, Dejing; Slezak, Dominik; Wang, Wei; Gruca, Aleksandra; Lin, Jerry Chun-Wei & Agrawal, Rakesh (Ed.), Proceedings 2023 IEEE International Conference on Big Data, Dec 15 - Dec 18, 2023, Sorrento, Italy. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798350324457. p. 4703–4711. doi: 10.1109/BigData59044.2023.10386277. Full text in Research Archive Show summary

  The exponential growth of data production emphasizes the importance of database management systems (DBMS) for managing vast amounts of data. However, the complexity of writing Structured Query Language (SQL) queries requires a diverse range of skills, which can be a challenge for many users. Different approaches are proposed to address this challenge by aiding SQL users in mitigating their skill gaps. One of these approaches is to design recommendation systems that provide several suggestions to users for writing their next SQL queries. Despite the availability of such recommendation systems, they often have several limitations, such as lacking sequence-awareness, session-awareness, and context-awareness. In this paper, we propose TRANSQLATION, a session-aware and sequence-aware recommendation system that recommends the fragments of the subsequent SQL query in a user session. We demonstrate that TRANSQLATION outperforms existing works by achieving, on average, 22% more recommendation accuracy when having a large amount of data and is still effective even when training data is limited. We further demonstrate that considering contextual similarity is a critical aspect that can enhance the accuracy and relevance of recommendations in query recommendation systems.

• Avram, Calin; Gligor, Adrian; Roman, Dumitru; Soylu, Ahmet; Nyulas, Victoria & Avram, Laura (2023). Machine learning based assessment of preclinical health questionnaires. International Journal of Medical Informatics. ISSN 1386-5056. 180. doi: 10.1016/j.ijmedinf.2023.105248. Full text in Research Archive Show summary

  Background: Within modern health systems, the possibility of accessing a large amount and a variety of data related to patients’ health has increased significantly over the years. The source of this data could be mobile and wearable electronic systems used in everyday life, and specialized medical devices. In this study we aim to investigate the use of modern Machine Learning (ML) techniques for preclinical health assessment based on data collected from questionnaires filled out by patients. Method: To identify the health conditions of pregnant women, we developed a questionnaire that was distributed in three maternity hospitals in the Mureș County, Romania. In this work we proposed and developed an ML model for pattern detection in common risk assessment based on data extracted from questionnaires. Results: Out of the 1278 women who answered the questionnaire, 381 smoked before pregnancy and only 216 quit smoking during the period in which they became pregnant. The performance of the model indicates the feasibility of the solution, with an accuracy of 98 % confirmed for the considered case study. Conclusion: The proposed solution offers a simple and efficient way to digitize questionnaire data and to analyze the data through a reduced computational effort, both in terms of memory and computing power used.

• Nikolov, Nikolay Vladimirov; Solberg, Arnor; Prodan, Radu; Soylu, Ahmet; Matskin, Mihhail & Roman, Dumitru (2023). Container-Based Data Pipelines on the Computing Continuum for Remote Patient Monitoring. Computer. ISSN 0018-9162. 56(10), p. 40–48. doi: 10.1109/MC.2023.3285414. Full text in Research Archive Show summary

  Diagnosing, treatment, and follow-up care of patients is happening increasingly through telemedicine, especially in remote areas where direct interaction is hindered. Over the past three years, following the COVID-19 pandemic, the utility of remote patient care has been further field-tested. Tackling the technical challenges of a growing demand for telemedicine requires a convergence of several fields: 1) software solutions for reliable, secure, and reusable data processing, 2) management of hardware resources (at scale) on the Cloud/Fog/Edge Computing Continuum, and 3) automation of DevOps processes for deployment of digital healthcare solutions with patients. In this context, the emerging concept of big data pipelines provides relevant solutions and is one of the main enablers. In what follows, we present a data pipeline for remote patient monitoring and show a real-world example of how data pipelines help address the stringent requirements of telemedicine.

• Bakken, Magnus & Soylu, Ahmet (2023). Chrontext: Portable SPARQL queries over contextualised time series data in industrial settings. Expert Systems With Applications. ISSN 0957-4174. 226. doi: 10.1016/j.eswa.2023.120149. Full text in Research Archive Show summary

  Industrial information models are standardised ways of representing industrial devices, equipment, and processes together with the data collected from associated sensors and control systems. Companies invest in such models to enable digitalisation and modular, reusable solutions. They also invest heavily in analytics (e.g. machine learning) based on time series data sets to improve operations. Queries that use such context to retrieve time series data can make industrial data sets more accessible to practitioners performing analytics and application development. Moreover, they can enable scalable deployment of resulting analytical models. Industrial availability constraints require that queries over context and time series should be portable in general, as they should be able retrieve data for training in a cloud setting and production data for deployment in an on-premise setting. Solving this problem is challenging with existing approaches as context and time series data tend to exist in separate, specialised databases. We address the issue by proposing a hybrid query engine, namely Chrontext, in the setting of a SPARQL database hosting the static model, and an arbitrary time series database. We show how with a set of annotations in the knowledge graph, SPARQL queries can be evaluated over such a hybrid architecture. We provide a proof showing that our approach correctly answers SPARQL 1.1. queries. We implement our approach in Rust under the Apache 2.0 license, and use the Apache Arrow-based Polars library together with configurable pushdowns to achieve high performance. We compare the performance of Chrontext against Ontop, one of the most prominent virtual knowledge graph systems, on a synthetic data set based on industrial standards. Data are stored in a S3 data lake and PostgreSQL with the Dremio data lakehouse as the SQL integrator. We find that our approach performs 10× to 85× faster and consumes much less memory than Ontop.

• Fatih, Gurcan; Menekse, Dalveren Gonca Gokce; Nergiz, Ercil Cagiltay & Soylu, Ahmet (2022). Detecting Latent Topics and Trends in Software Engineering Research Since 1980 Using Probabilistic Topic Modeling. IEEE Access. 10, p. 74638–74654. doi: 10.1109/ACCESS.2022.3190632. Full text in Research Archive Show summary

  The landscape of software engineering research has changed significantly from one year to the next in line with industrial needs and trends. Therefore, today’s research literature on software engineering has a rich and multidisciplinary content that includes a large number of studies; however, not many of them demonstrate a holistic view of the field. From this perspective, this study aimed to reveal a holistic view that reflects topics, trends, and trajectories in software engineering research by analyzing the majority of domain-specific articles published over the last 40 years. This study first presents an objective and systematic method for corpus creation through major publication sources in the field. A corpus was then created using this method, which includes 44 domain-specific conferences and journals and 57,174 articles published between 1980 and 2019. Next, this corpus was analyzed using an automated text-mining methodology based on a probabilistic topic-modeling approach. As a result of this analysis, 24 main topics were found. In addition, topical trends in the field were revealed. Finally, three main developmental stages of the field were identified as: the programming age, the software development age, and the software optimization age.

• Gurcan, Fatih; Dalveren, Gonca Gokce Menekse; Cagiltay, Nergiz Ercil; Roman, Dumitru & Soylu, Ahmet (2022). Evolution of Software Testing Strategies and Trends: Semantic Content Analysis of Software Research Corpus of the Last 40 Years. IEEE Access. 10, p. 106093–106109. doi: 10.1109/ACCESS.2022.3211949. Full text in Research Archive Show summary

  From the early days of computer systems to the present, software testing has been considered as a crucial process that directly affects the quality and reliability of software-oriented products and services. Accordingly, there is a huge amount of literature regarding the improvement of software testing approaches. However, there are limited reviews that show the whole picture of the software testing studies covering the topics and trends of the field. This study aims to provide a general figure reflecting topics and trends of software testing by analyzing the majority of software testing articles published in the last 40 years. A semi-automated methodology is developed for the analysis of software testing corpus created from core publication sources. The methodology of the study is based on the implementation of probabilistic topic modeling approach to discover hidden semantic patterns in the 14,684 published articles addressing software testing issues between 1980 and 2019. The results revealed 42 topics of the field, highlighting five software development ages, namely specification, detection, generation, evaluation, and prediction. The recent accelerations of the topics also showed a trend toward prediction-based software testing actions. Additionally, a higher trend on the topics concerning “Security Vulnerability”, “Open Source” and “Mobile Application” was identified. This study showed that the current trend of software testing is towards prediction-based testing strategies. Therefore, the findings of this study may provide valuable insights for the industry and software communities to be prepared for the possible changes in the software testing procedures using prediction-based approaches.

• Roman, Dumitru; Prodan, Radu; Nikolov, Nikolay; Soylu, Ahmet; Matskin, Mihhail & Marrella, Andrea [Show all 13 contributors for this article] (2022). Big Data Pipelines on the Computing Continuum: Tapping the Dark Data. Computer. ISSN 0018-9162. 55(11), p. 74–84. doi: 10.1109/MC.2022.3154148. Full text in Research Archive Show summary

  Big Data pipelines are essential for leveraging Dark Data, i.e., data collected but not used and turned into value. However, tapping their potential requires going beyond existing approaches and frameworks for Big Data processing. The Computing Continuum enables new opportunities for managing Big Data pipelines concerning efficient management of heterogeneous and untrustworthy resources. This article discusses the Big Data pipelines lifecycle on the Computing Continuum, its associated challenges and outlines a future research agenda in this area.

• Soylu, Ahmet; Corcho, Oscar; Elvesæter, Brian; Badenes-Olmedo, Carlos; Martínez, Francisco Yedro & Kovavic, Matej [Show all 13 contributors for this article] (2022). Data Quality Barriers for Transparency in Public Procurement. Information. 13(2). doi: 10.3390/info13020099. Full text in Research Archive Show summary

  Governments need to be accountable and transparent for their public spending decisions in order to prevent losses through fraud and corruption as well as to build healthy and sustainable economies. Open data act as a major instrument in this respect by enabling public administrations, service providers, data journalists, transparency activists, and regular citizens to identify fraud or uncompetitive markets through connecting related, heterogeneous, and originally unconnected data sources. To this end, in this article, we present our experience in the case of Slovenia, where we successfully applied a number of anomaly detection techniques over a set of open disparate data sets integrated into a Knowledge Graph, including procurement, company, and spending data, through a linked data-based platform called TheyBuyForYou. We then report a set of guidelines for publishing high quality procurement data for better procurement analytics, since our experience has shown us that there are significant shortcomings in the quality of data being published. This article contributes to enhanced policy making by guiding public administrations at local, regional, and national levels on how to improve the way they publish and use procurement-related data; developing technologies and solutions that buyers in the public and private sectors can use and adapt to become more transparent, make markets more competitive, and reduce waste and fraud; and providing a Knowledge Graph, which is a data resource that is designed to facilitate integration across multiple data silos by showing how it adds context and domain knowledge to machine-learning-based procurement analytic.

• Soylu, Ahmet; Corcho, Oscar; Elvesæter, Brian; Badenes-Olmedo, Carlos; Blount, Tom & Martínez, Francisco Yedro [Show all 13 contributors for this article] (2022). TheyBuyForYou platform and knowledge graph: Expanding horizons in public procurement with open linked data. Semantic Web – Interoperability, Usability, Applicability. ISSN 1570-0844. 13(2), p. 265–291. doi: 10.3233/SW-210442. Full text in Research Archive Show summary

  Public procurement is a large market affecting almost every organisation and individual; therefore, governments need to ensure its efficiency, transparency, and accountability, while creating healthy, competitive, and vibrant economies. In this context, open data initiatives and integration of data from multiple sources across national borders could transform the procurement market by such as lowering the barriers of entry for smaller suppliers and encouraging healthier competition, in particular by enabling cross-border bids. Increasingly more open data is published in the public sector; however, these are created and maintained in siloes and are not straightforward to reuse or maintain because of technical heterogeneity, lack of quality, insufficient metadata, or missing links to related domains. To this end, we developed an open linked data platform, called TheyBuyForYou, consisting of a set of modular APIs and ontologies to publish, curate, integrate, analyse, and visualise an EU-wide, cross-border, and cross-lingual procurement knowledge graph. We developed advanced tools and services on top of the knowledge graph for anomaly detection, cross-lingual document search, and data storytelling. This article describes the TheyBuyForYou platform and knowledge graph, reports their adoption by different stakeholders and challenges and experiences we went through while creating them, and demonstrates the usefulness of Semantic Web and Linked Data technologies for enhancing public procurement.

• Zhou, Dongzhuoran; Zhou, Baifan; Zheng, Zhuoxun; Soylu, Ahmet; Savkovic, Ognjen & Kostylev, Egor [Show all 7 contributors for this article] (2022). ScheRe: Schema Reshaping for Enhancing Knowledge Graph Construction. In Hasan, Mohammad Al & Xiong, Li (Ed.), CIKM'22: Proceedings of the 31st ACM International Conference on Information & Knowledge Management. Association for Computing Machinery (ACM). ISSN 9781450392365. p. 5074–5078. doi: 10.1145/3511808.3557214. Full text in Research Archive Show summary

  Automatic knowledge graph (KG) construction is widely used for e.g. data integration, question answering and semantic search. There are many approaches of automatic KG construction. Among which, an important approach is to map the raw data to a given domain KG schema, e.g., domain ontology or conceptual graph, and construct the entities and properties according to the domain KG schema. However, the existing approaches to construct KGs are not always efficient enough and the resulting KGs are not sufficiently user-friendly. The main challenge arises from the trade-off: the domain KG schema should be knowledge-oriented, to reflect the general domain knowledge; while a KG schema should be dataoriented, to cover all data features. If the former is directly used for KG construction, this can cause issues like a high load of blank nodes, which are technical nodes in the KGs that represent unknown entities. To this end, we propose our ScheRe system in the demo, which relies on a schema reshaping algorithm and other two semantic modules for enhancing KG construction. The demo attendees will use ScheRe to reshape a domain KG schema to data specific KG schema, build KGs with industrial data, and experience more user-friendly querying.

• Roman, Dumitru; Alexiev, Vladimir; Paniagua, Javier; Elvesæter, Brian; Zernichow, Bjørn Marius von & Soylu, Ahmet [Show all 8 contributors for this article] (2022). The euBusinessGraph ontology: A lightweight ontology for harmonizing basic company information. Semantic Web – Interoperability, Usability, Applicability. ISSN 1570-0844. 13(1), p. 41–68. doi: 10.3233/SW-210424. Full text in Research Archive Show summary

  Company data, ranging from basic company information such as company name(s) and incorporation date to complex balance sheets and personal data about directors and shareholders, are the foundation that many data value chains depend upon in various sectors (e.g., business information, marketing and sales, etc.). Company data becomes a valuable asset when data is collected and integrated from a variety of sources, both authoritative (e.g., national business registers) and non-authoritative (e.g., company websites). Company data integration is however a difficult task primarily due to the heterogeneity and complexity of company data, and the lack of generally agreed upon semantic descriptions of the concepts in this domain. In this article, we introduce the euBusinessGraph ontology as a lightweight mechanism for harmonising company data for the purpose of aggregating, linking, provisioning and analysing basic company data. The article provides an overview of the related work, ontology scope, ontology development process, explanations of core concepts and relationships, and the implementation of the ontology. Furthermore, we present scenarios where the ontology was used, among others, for publishing company data (business knowledge graph) and for comparing data from various company data providers. The euBusinessGraph ontology serves as an asset not only for enabling various tasks related to company data but also on which various extensions can be built upon.

• Islam, Md Nazmul; Hasan, Mehedi; Hossain, Md Kabir; Alam, Md Golam Rabiul; Uddin, Md Zia & Soylu, Ahmet (2022). Vision transformer and explainable transfer learning models for auto detection of kidney cyst, stone and tumor from CT-radiography. Scientific Reports. 12(1). doi: 10.1038/s41598-022-15634-4. Full text in Research Archive Show summary

  Renal failure, a public health concern, and the scarcity of nephrologists around the globe have necessitated the development of an AI-based system to auto-diagnose kidney diseases. This research deals with the three major renal diseases categories: kidney stones, cysts, and tumors, and gathered and annotated a total of 12,446 CT whole abdomen and urogram images in order to construct an AI-based kidney diseases diagnostic system and contribute to the AI community’s research scope e.g., modeling digital-twin of renal functions. The collected images were exposed to exploratory data analysis, which revealed that the images from all of the classes had the same type of mean color distribution. Furthermore, six machine learning models were built, three of which are based on the state-of-the-art variants of the Vision transformers EANet, CCT, and Swin transformers, while the other three are based on well-known deep learning models Resnet, VGG16, and Inception v3, which were adjusted in the last layers. While the VGG16 and CCT models performed admirably, the swin transformer outperformed all of them in terms of accuracy, with an accuracy of 99.30 percent. The F1 score and precision and recall comparison reveal that the Swin transformer outperforms all other models and that it is the quickest to train. The study also revealed the blackbox of the VGG16, Resnet50, and Inception models, demonstrating that VGG16 is superior than Resnet50 and Inceptionv3 in terms of monitoring the necessary anatomy abnormalities. We believe that the superior accuracy of our Swin transformer-based model and the VGG16-based model can both be useful in diagnosing kidney tumors, cysts, and stones.

• Roman, Dumitru; Pultier, Antoine; Ma, Xiang; Soylu, Ahmet & Ulyashin, Alexander (2022). Data quality issues in solar panels installations: a case study. In Nguyen, Phu Hong; Magnanini, Maria Chiara & Sen, Sagar (Ed.), Proceedings of the 2nd International Workshop on Software Engineering and AI for Data Quality in Cyber-Physical Systems/Internet of Things (SEA4DQ 2022). Association for Computing Machinery (ACM). ISSN 9781450394598. p. 24–25. doi: 10.1145/3549037.3564120. Full text in Research Archive Show summary

  Solar photovoltaics (PV) is becoming an important source of global electricity generation. Modern PV installations come with a variety of sensors attached to them for monitoring purposes (e.g., maintenance, prediction of electricity generation, etc.). Data collection (and implicitly the quality of data) from PV systems is becoming essential in this context. In this position paper, we introduce a modern PV mini power plant demo site setup for research purposes and discuss the data quality issues we encountered in operating the power plant.

• Zheng, Zhuoxun; Zhou, Baifan; Soylu, Ahmet & Kharlamov, Evgeny (2022). Towards a Visualisation Ontology for Data Analysis in Industrial Applications. CEUR Workshop Proceedings. 3355. Full text in Research Archive Show summary

  Machine learning (ML) approaches have proven their great potential in dealing with heterogeneous and voluminous data and thus are widespread in industry. To facilitate the presentation of the ML results and the subsequent discussion on that, visualisation is essential, as it effectively conveys the information behind the data. However, a standardisation of the knowledge and practice about visualisation is still lacking in the industry, which sometimes leads to misunderstandings in conveying information and thus making the discussions on the ML results error-prone. A visualisation ontology which models the nature and pipeline of visualization tasks are well suited to provide such standardisation. Currently there are a few studies that discuss partially the modelling of visualisation, however they are less adequate in depicting the practical procedure of visualisation tasks, which is highly demanded in the industrial applications. To this end, we present our ongoing work of development of the visualisation ontology in industrial applications at Bosch. We also discuss applications and evaluation of our ontology.

• Song, Hui; Soylu, Ahmet & Roman, Dumitru (2022). Towards Cognitive Self-Management of IoT-Edge-Cloud Continuum based on User Intents. In Zhao, Xinghui & Sill, Alan (Ed.), 15th IEEE/ACM International Conference on Utility and Cloud Computing. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665460873. p. 313–316. doi: 10.1109/UCC56403.2022.00055. Full text in Research Archive Show summary

  Elasticity of the computing continuum with on demand availability allows for automated provisioning and release of computing resources as needed; however, this self management capability is severely limited due to the lack of knowledge on historical and timely resource utilisation and means for stakeholders to express their needs in a high-level manner. In this paper, we introduce and discuss a new concept – intent-based cognitive continuum for sustainable elasticity.

• Khan, Akif Quddus; Nikolov, Nikolay Vladimirov; Matskin, Mihhail; Prodan, Radu; Song, Hui & Roman, Dumitru [Show all 7 contributors for this article] (2022). Smart Data Placement for Big Data Pipelines: An Approach based on the Storage-as-a-Service Model. In Zhao, Xinghui & Sill, Alan (Ed.), 15th IEEE/ACM International Conference on Utility and Cloud Computing. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665460873. p. 317–320. doi: 10.1109/UCC56403.2022.00056. Full text in Research Archive Show summary

  The development of big data pipelines is a challenging task, especially when data storage is considered as part of the data pipelines. Local storage is expensive, hard to maintain, comes with several challenges (e.g., data availability, data security, and backup). The use of cloud storage, i.e., Storageas-a-Service (StaaS), instead of local storage has the potential of providing more flexibility in terms of such as scalability, fault tolerance, and availability. In this paper, we propose a generic approach to integrate StaaS with data pipelines, i.e., computation on an on-premise server or on a specific cloud, but integration with StaaS, and develop a ranking method for available storage options based on five key parameters: cost, proximity, network performance, the impact of server-side encryption, and user weights. The evaluation carried out demonstrates the effectiveness of the proposed approach in terms of data transfer performance and the feasibility of dynamic selection of a storage option based on four primary user scenarios.

• Thomas, Aleena; Nikolov, Nikolay; Pultier, Antoine; Roman, Dumitru; Elvesæter, Brian & Soylu, Ahmet (2022). SIM-PIPE DryRunner: An approach for testing container-based big data pipelines and generating simulation data. Computer Software and Applications Conference. ISSN 0730-3157. p. 1159–1164. doi: 10.1109/COMPSAC54236.2022.00182. Full text in Research Archive Show summary

  Big data pipelines are becoming increasingly vital in a wide range of data intensive application domains such as digital healthcare, telecommunication, and manufacturing for efficiently processing data. Data pipelines in such domains are complex and dynamic and involve a number of data processing steps that are deployed on heterogeneous computing resources under the realm of the Edge-Cloud paradigm. The processes of testing and simulating big data pipelines on heterogeneous resources need to be able to accurately represent this complexity. However, since big data processing is heavily resource-intensive, it makes testing and simulation based on historical execution data impractical. In this paper, we introduce the SIM - PIPE Dry Runner approach - a dry run approach that deploys a big data pipeline step by step in an isolated environment and executes it with sample data; this approach could be used for testing big data pipelines and realising practical simulations using existing simulators.

• Lin, Tengteng; Chen, Qiaosheng; Cheng, Gong; Soylu, Ahmet; Ell, Basil & Zhao, Ruoqi [Show all 10 contributors for this article] (2022). ACORDAR: A Test Collection for Ad Hoc Content-Based (RDF) Dataset Retrieval. In Amigó, Enrique; Castells, Pablo; Gonzalo, Julio; Carterette, Ben; Culpepper, J. Shane & Kazai, Gabriella (Ed.), SIGIR '22: The 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, Madrid, Spain, July 11 - 15, 2022. Association for Computing Machinery (ACM). ISSN 9781450387323. p. 2981–2991. doi: 10.1145/3477495.3531729. Full text in Research Archive Show summary

  Ad hoc dataset retrieval is a trending topic in IR research. Methods and systems are evolving from metadata-based to content-based ones which exploit the data itself for improving retrieval accuracy but thus far lack a specialized test collection. In this paper, we build and release the first test collection for ad hoc content-based dataset retrieval, where content-oriented dataset queries and content-based relevance judgments are annotated by human experts who are assisted with a dashboard designed specifically for comprehensively and conveniently browsing both the metadata and data of a dataset. We conduct extensive experiments on the test collection to analyze its difficulty and provide insights into the underlying task.

• Tahmasebi, Shirin; Payberah, Amir Hossein; Soylu, Ahmet; Roman, Dumitru & Matskin, Mihhail (2022). TranSQL: A Transformer-based Model for Classifying SQL Queries. In IEEE, . (Eds.), Proceedings 21st IEEE international conference on machine learning and applications : ICMLA 2022. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665462839. p. 788–793. doi: 10.1109/ICMLA55696.2022.00131. Full text in Research Archive Show summary

  Domain-Specific Languages (DSL) are becoming popular in various fields as they enable domain experts to focus on domain-specific concepts rather than software-specific ones. Many domain experts usually reuse their previously-written scripts for writing new ones; however, to make this process straightforward, there is a need for techniques that can enable domain experts to find existing relevant scripts easily. One fundamental component of such a technique is a model for identifying similar DSL scripts. Nevertheless, the inherent nature of DSLs and lack of data makes building such a model challenging. Hence, in this work, we propose TRANSQL, a transformer-based model for classifying DSL scripts based on their similarities, considering their few-shot context. We build TRANSQL using BERT and GPT-3, two performant language models. Our experiments focus on SQL as one of the most commonly-used DSLs. The experiment results reveal that the BERT-based TRANSQL cannot perform well for DSLs since they need extensive data for the fine-tuning phase. However, the GPT-based TRANSQL gives markedly better and more promising results.

• Zheng, Zhuoxun; Zhou, Baifan; Zhou, Dongzhuoran; Soylu, Ahmet & Kharlamov, Evgeny (2022). ExeKG: Executable Knowledge Graph System for User-friendly Data Analytics. In Hasan, Mohammad Al & Xiong, Li (Ed.), CIKM'22: Proceedings of the 31st ACM International Conference on Information & Knowledge Management. Association for Computing Machinery (ACM). ISSN 9781450392365. p. 5064–5068. doi: 10.1145/3511808.3557195. Full text in Research Archive Show summary

  Data analytics including machine learning (ML) is essential to extract insights from production data in modern industries. However, industrial ML is affected by: the low transparency of ML towards non-ML experts; poor and non-unified descriptions of ML practices for reviewing or comprehension; ad-hoc fashion of ML solutions tailored to specific applications, which affects their re-usability. To address these challenges, we propose the concept and a system of executable knowledge graph (KG), which represent KGs that rely on semantic technologies to formally encode ML knowledge and solutions. These KGs can be translated to executable scripts in a reusable and modularised fashion.

• Roman, Dumitru; Song, Hui; Loupos, Konstantinos; Krousarlis, Thomas; Soylu, Ahmet & Skarmeta, Antonio F. (2022). The Computing Fleet: Managing Microservices-based Applications on the Computing Continuum, 2022 IEEE 19th International Conference on Software Architecture Companion (ICSA-C). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665494939. p. 40–44. doi: 10.1109/ICSA-C54293.2022.00015. Full text in Research Archive Show summary

  In this paper we propose the concept of "Computing Fleet" as an abstract entity representing groups of heterogeneous, distributed, and dynamic infrastructure elements across the Computing Continuum (covering the Edge- Fog-Cloud computing paradigms). In the process of using fleets, stakeholders obtain the virtual resources from the fleet, deploy software applications to the fleet, and control the data flow, without worrying about what devices are used in the fleet, how they are connected, and when they may join and exit the fleet. We propose a three-layer reference architecture for the Computing Fleet capturing key elements for designing and operating fleets. We discuss key aspects related to the management of microservices-based applications on the Computing Fleet and propose an approach for deployment and orchestration of microservices-based applications on fleets. Furthermore, we present a software prototype as a preliminary evaluation of the Computing Fleet concept in a concrete Cloud- Edge scenario related to remote patients monitoring.

View all works in NVA

• Soylu, Ahmet; Liu, Fan; Mitra, Karan; Zhang, Yan & Grønli, Tor-Morten (2026). Mobile and Ubiquitous Systems: Computing, Networking and Services - 21st EAI International Conference, MobiQuitous 2024, Oslo, Norway, November 12–14, 2024, Proceedings. Springer Nature. ISBN 9783032105547. 633 p. doi: https:/link.springer.com/book/10.1007/978-3-032-10554-7. Full text in Research Archive
• Hogan, Aidan; Satoh, Ken; Dağ, Hasan; Turhan, Anni-Yasmin; Roman, Dumitru & Soylu, Ahmet (2025). Rules and Reasoning: 9th International Joint Conference, RuleML+RR 2025, Istanbul, Turkey, September 22–24, 2025, Proceedings. Springer. ISBN 9783032088871. 258 p. doi: 10.1007/978-3-032-08887-1. Full text in Research Archive Show summary

  RuleML+RR focusses on theoretical advances, novel technologies, and innovative applications relating to knowledge representation and reasoning with rules.

• Rula, Anisa; Sallinger, Emanuel; Savkovic, Ognjen; Ciuciu, Ioana Georgiana; Toma, Ioan & Parreira, Josiane Xavier [Show all 9 contributors for this article] (2024). Companion Proceedings of the 8th International Joint Conference on Rules and Reasoning co-located with 20th Reasoning Web Summer School (RW 2024) and 16th DecisionCAMP 2024 as part of Declarative AI 2024. CEUR Workshop Proceedings. ISBN 9783031724060. doi: https:/ceur-ws.org/Vol-3816/. Full text in Research Archive
• Fensel, Anna; Ozaki, Ana; Dumitru, Roman & Soylu, Ahmet (2023). Rules and Reasoning - 7th International Joint Conference, RuleML+RR 2023, Oslo, Norway, September 18-20, 2023, Proceedings. Springer Publishing Company. ISBN 9783031450716. 269 p. doi: 10.1007/978-3-031-45072-3. Full text in Research Archive

View all works in NVA

• Soylu, Ahmet (2026). En ph.d. bør være selvstendig, ikke isolert. doi: https:/www.universitetsavisa.no/doktorgradsutdanning-forskningsmiljo-phd/en-phd-bor-vaere-selvstendig-ikke-isolert/458789. Full text in Research Archive Show summary

  - Jo mer jeg har sett doktorgradsutdanning fra ulike sider, desto mer overbevist har jeg blitt om én ting: En ph.d. må være selvstendig arbeid, men den bør ikke være isolert arbeid.

• Soylu, Ahmet (2026). Vi trenger ikke færre doktorgrader, men et mer mangfoldig system. doi: https:/www.forskerforum.no/innsendt/vi-trenger-ikke-faerre-doktorgrader-men-mer-mangfold/. Full text in Research Archive Show summary

  Problemet er ikke antallet personer med doktorgrad, men den snevre måten vi organiserer doktorgradsutdanningen.

• Soylu, Ahmet (2026). Ekte tilhørighet oppsto tett på mennesker. doi: https:/www.samtiden.no/essay/ekte-tilhorighet-oppsto-tett-pa-mennesker/1411868. Full text in Research Archive Show summary

  Utforsk hvordan likestilling og mangfold utfordres i Norge gjennom personlige erfaringer med tilhørighet og kulturell forståelse.

• Soylu, Ahmet (2026). Når universitetet møter etiske dilemmaer. doi: https:/www.forskerforum.no/innsendt/uio-debatten-handler-om-mer-enn-boikott/. Full text in Research Archive Show summary

  Debatten om akademisk boikott ved Universitetet i Oslo (UiO) reiser en bredere problemstilling enn selve vedtaket: hvordan universiteter bør resonnere når sentrale verdier trekker i ulike retninger, skriver professor Ahmet Soylu.

• Soylu, Ahmet (2026). KI er kommet for å bli. doi: https:/www.universitetsavisa.no/forskning-ki-verktoy-kunstig-intelligens/ki-er-kommet-for-a-bli/456121. Full text in Research Archive Show summary

  Verden står i en fornuftskrise, men ikke på grunn av KI eller annen teknologi. Åpne en hvilken som helst avis. Overskriftene taler for seg selv.

• Soylu, Ahmet (2026). Akademias usynlige murer. doi: https:/www.khrono.no/akademias-usynlige-murer/1035201. Full text in Research Archive Show summary

  Debatten om administrasjon og akademisk frihet peker egentlig mot et enda dypere problem, mener dekan Ahmet Soylu.

• Soylu, Ahmet (2026). Er vi imot verktøybruk som sådan? doi: https:/agendamagasin.no/debatt/er-vi-imot-verktoybruk-som-sadan/. Full text in Research Archive Show summary

  Mennesker har alltid brukt verktøy. Med nye verktøy blir også mer automatisert. En verden uten verktøy er ikke et seriøst alternativ.

• Soylu, Ahmet (2026). Ja, jeg bruker KI i debattinnlegg. doi: https:/www.aftenposten.no/meninger/debatt/i/y5RQya/ja-jeg-bruker-ki-i-debattinnlegg. Full text in Research Archive Show summary

  Jeg har tenkt mye på hvor grensene bør gå, skriver Ahmet Soylu.

• Soylu, Ahmet (2026). Universitetsadministrasjon er ikke en støttefunksjon. doi: https:/www.khrono.no/universitetsadministrasjon-er-ikke-en-stottefunksjon/1028017. Full text in Research Archive Show summary

  Et universitet kan være fullstendig administrerbart, og likevel gradvis miste evnen til å styre seg selv. Forskjellen ligger i om administrasjonen behandles som en kostnad man kutter, eller som institusjonelt forvalterskap.

• Soylu, Ahmet (2025). Finnes det noe hellig igjen, noe vi bøyer oss for? doi: https:/www.samtiden.no/essay/finnes-det-noe-hellig-igjen-noe-vi-boyer-oss-for/1331739. Full text in Research Archive Show summary

  Vi løfter historiefortelling over oppriktighet, strategi over sjel. I denne nye ordenen bøyer vi oss ikke lenger, ikke engang i overført betydning. Ærefrykt har ingen markedsverdi.

• Soylu, Ahmet (2025). Høyere utdanning trenger en sjel. doi: https:/www.forskerforum.no/innsendt/hoyere-utdanning-trenger-en-sjel/. Full text in Research Archive Show summary

  Akademikere kan ikke ledes med makt eller tall alene – de trenger tid, rom og frihet til å skape, skriver Ahmet Soylu.

• Soylu, Ahmet (2025). Utenfor boblen: Hva skylder vi verden? doi: https:/www.forskerforum.no/innsendt/utenfor-boblen-hva-skylder-vi-verden/. Full text in Research Archive Show summary

  Norge har potensial til å bli et hjem for internasjonale hoder og samvittigheter. Og vi med utenlandsk bakgrunn må delta mer i den offentlige samtalen, skriver Ahmet Soylu.

• Soylu, Ahmet (2025). KI og alderen for sterile problemer. doi: https:/www.forskerforum.no/innsendt/kunstig-intelligens-er-ikke-var-krise-den-speiler-bare-det-vi-har-blitt/. Full text in Research Archive Show summary

  Den virkelige faren er ikke at smartere maskiner tar over, men at mennesker og samfunn uten verdier gjør det.

• Soylu, Ahmet (2025). Å tale er å føle og tenke. doi: https:/www.khrono.no/a-tale-er-a-fole-og-tenke/985251. Full text in Research Archive Show summary

  Språk handler ikke bare om ord, men om mening. Det svekkes når vi mister evnen eller viljen til å føle og tenke i dybden, ikke når vi bruker et annet språk, skriver Ahmet Soylu.

• Soylu, Ahmet (2025). Akademia mellom idealisme og pragmatisme. doi: https:/www.khrono.no/akademia-mellom-idealisme-og-pragmatisme/963025. Full text in Research Archive Show summary

  Akademikere over hele verden jobber mer enn de får betalt for — i ukedager, ferier, dag og natt. Tenkning er ikke noe man kan legge igjen på kontoret og hente opp igjen neste arbeidsdag. Den følger deg. Den lever i deg.

• Soylu, Ahmet (2025). Fabula acta est. doi: https:/www.khrono.no/fabula-acta-est/1002643. Full text in Research Archive Show summary

  Universiteter er mer enn produksjonslinjer; de er steder for tilhørighet, skapelse og ansvar. Likevel truer sterk politisk styring og tomme prestisjebekymringer med å redusere forskning til rutine.

• Zheng, Zhuoxun; Ognjen, Savković,; Zhou, Baifan; Antonis, Klironomos,; Kharlamov, Evgeny & Soylu, Ahmet (2025). Graph Constraint Language for Industrial Knowledge Graphs and Machine Learning. doi: 10.1007/978-3-032-02215-8_13. Full text in Research Archive Show summary

  Knowledge graphs (KGs) are seeing widespread application across diverse domains, including querying, search, machine learning (ML), and stream data processing. In all these areas, ensuring data consistency and quality is of critical importance. For instance,...

• Avogadro, Roberto; Rauniyar, Ashish; Goknil, Arda; Soylu, Ahmet; Udroiu, Cosmin & Ma, Xiang [Show all 7 contributors for this article] (2025). Human-in-the-Loop Entity Linking Visualization and Refinement with KOALA-UI and the Crocodile Framework. Full text in Research Archive
• Jayawardene, Iroshani; Skrzycki, Magdalena; Ma, Xiang; Soylu, Ahmet & Roman, Dumitru (2025). Synthetic Data Generation for Electric Vehicle Charging. Full text in Research Archive
• Avogadro, Roberto; Jayawardene, Iroshani; Ma, Xiang; Soylu, Ahmet & Roman, Dumitru (2024). Koala-UI: An Interactive User Interface for Tabular Data Linking. Full text in Research Archive
• Jayawardene, Iroshani; Roman, Dumitru; Zhao, Yingxiang; Ulyashin, Alexsander G.; Soylu, Ahmet & Ma, Xiang (2024). Towards an Open Energy Management System for Integrated Energy Storage and Electric Vehicle Fast Charging Station. Full text in Research Archive
• Kliegr, Tomáš; Gutierrez-Basulto, Victor & Soylu, Ahmet (2023). Introduction to the Special Issue on Logic Rules and Reasoning: Selected Papers from the 4th International Joint Conference on Rules and Reasoning (RuleML+RR 2020). Theory and Practice of Logic Programming. ISSN 1471-0684. p. 1–4. doi: 10.1017/S1471068423000042. Full text in Research Archive
• Zheng, Zhuoxun; Zhou, Baifan; Zhou, Dongzhuoran; Khan, Akif Quddus; Soylu, Ahmet & Kharlamov, Evgeny (2022). Towards A Statistic Ontology for Data Analysis in Smart Manufacturing. Full text in Research Archive
• Zheng, Zhuoxun; Zhou, Baifan; Zhou, Dongzhuoran; Soylu, Ahmet & Kharlamov, Evgeny (2022). Executable Knowledge Graph for Transparent Machine Learning in Welding Monitoring at Bosch. In Hasan, Mohammad Al & Xiong, Li (Ed.), CIKM'22: Proceedings of the 31st ACM International Conference on Information & Knowledge Management. Association for Computing Machinery (ACM). ISSN 9781450392365. p. 5102–5103. doi: 10.1145/3511808.3557512. Full text in Research Archive Show summary

  With the development of Industry 4.0 technology, modern industries such as Bosch’s welding monitoring witnessed the rapid widespread of machine learning (ML) based data analytical applications, which in the case of welding monitoring has led to more efficient and accurate welding monitoring quality. However, industrial ML is affected by the low transparency of ML towards non-ML experts needs. The lack of understanding by domain experts of ML methods hampers the application of ML methods in industry and the reuse of developed ML pipelines, as ML methods are often developed in an ad hoc manner for specific problems. To address these challenges, we propose the concept and a system of executable Knowledge Graph (KG), which formally encode ML knowledge and solutions in KGs, which serve as common language between ML experts and nonML experts, thus facilitate their communication and increase the transparency of ML methods. We evaluated our system extensively with an industrial use case at Bosch, showing promising results.

• Roman, Dumitru; Pultier, Antoine; Ma, Xiang; Soylu, Ahmet & Ulyashin, Alexsander G. (2022). Data quality issues in solar panels installations: a case study. Full text in Research Archive
• Zhou, Dongzhuoran; Zhou, Baifan; Zheng, Zhuoxun; Soylu, Ahmet; Cheng, Gong & Jimenez-Ruiz, Ernesto [Show all 8 contributors for this article] (2022). Ontology Reshaping for Knowledge Graph Construction: Applied on Bosch Welding Case. Full text in Research Archive
• Zheng, Zhuoxun; Zhou, Baifan; Zhou, Dongzhuoran; Cheng, Gong; Jimenez-Ruiz, Ernesto & Soylu, Ahmet [Show all 7 contributors for this article] (2022). Executable Knowledge Graphs for Machine Learning: A Bosch Case of Welding Monitoring. Full text in Research Archive

View all works in NVA