Search Results (234)

This work addresses the need for efficient and reusable vector embeddings (VEs) for large-scale RDF Knowledge Graphs, focusing on the widely used DrugBankdataset. The study aims to reduce the computational burden and environmental impact associated with repeatedly generating embeddings for downstream tasks such as link prediction, clustering, and recommendation. To this end, embeddings are generated from the DrugBank Knowledge Graph—comprising over 3.6 million triples, more than 1.5 million entities, and 95 relations—using 25 models implemented in the PyKEEN framework. The dataset is processed into training, validation, and test splits, and embeddings of fixed dimensionality are produced for both entities and relations. The resulting representations are released in multiple formats, including full JSON files, class-partitioned subsets, and an efficient Parquet-based structure, to support scalable querying. Experimental evaluation demonstrates substantial improvements in access time and memory usage when using structured formats, particularly Parquet. Additionally, the study quantifies the carbon footprint of embedding generation, showing that distributing precomputed embeddings can reduce energy consumption by up to 99.37% compared to on-demand recomputation. Overall, the work provides a comprehensive, reusable resource that facilitates research while promoting computational efficiency and environmental sustainability. Full article

Background: Perioperative antithrombotic management is a complex and high-risk aspect of anesthesia practice, requiring a careful balance between bleeding and thromboembolic risks. Large language models (LLMs) are increasingly explored for clinical decision support, but their reliability in complex perioperative antithrombotic scenarios remains uncertain. Methods: In this blinded, scenario-based study, 100 elective surgical cases, including grey-zone scenarios, were developed to reflect clinically relevant perioperative antithrombotic decisions. The European Society of Anaesthesiology and Intensive Care/European Society of Regional Anesthesia and Pain Therapy (ESAIC/ESRA) 2022 guideline was incorporated into the standardized prompt as a common framework for neuraxial safety and relevant antithrombotic interruption intervals. Four LLMs (GPT-5.4 Thinking, Claude Opus 4.5, DeepSeek v3.2 Reasoning and Qwen3.5-Plus) were evaluated using a standardized prompt. Model responses were presented without model identity and independently assessed by two experienced anesthesiologists, neither of whom was an author of the present study, across five clinical domains on a 5-point Likert scale, with a separate clinical applicability assessment. Results: Using a uniform four-domain composite across all 100 scenarios, overall expert-rated performance differed across models (Friedman χ² = 28.75, p < 0.001; Kendall’s W = 0.096). Claude had the highest median primary composite score, although its difference from DeepSeek was not statistically significant. Leave-one-domain-out sensitivity analyses showed that between-model separation was particularly sensitive to the clinical-rationale domain; exclusion of this domain reduced Kendall’s W to 0.030. A secondary five-domain analysis restricted to the 51 common-proceed scenarios, all of which were standard rather than grey-zone cases, also showed an overall between-model difference (Friedman χ² = 55.13, p < 0.001; Kendall’s W = 0.360). Because each scenario was queried only once, fine-grained differences in model ordering should be considered provisional. Although all models performed well in drug discontinuation decisions, greater variability was observed in discontinuation timing, anesthesia choice and clinical reasoning. Postponement behaviour varied substantially across models, with DeepSeek and Qwen recommending postponement more often than ChatGPT and Claude. Conclusions: LLMs show promise as supportive tools in perioperative decision-making, but their performance remains variable, especially in complex situations. At present, they should be used cautiously and always under expert supervision, rather than as independent decision-makers. Full article

Central Asian populations remain underrepresented in pharmacogenomic research, limiting the availability of population-specific data for genotype-informed prescribing and precision medicine. This study analyzed clinically relevant pharmacogenomic variant frequencies in Kazakh, Russian, and Uzbek population groups residing in Kazakhstan using genome-wide genotype data from 1301 individuals: Kazakh (n = 1111), Russian (n = 156), and Uzbek (n = 34). ClinPGx, a PharmGKB-based clinical annotation framework that prioritizes variant–drug associations according to levels of evidence, was used to select variants with evidence levels 1A, 1B, and 2A. In total, 112 directly genotyped variants were retained for population-specific allele and genotype frequency analysis. All 112 variants were queried against the gnomAD v4.1 genome and exome reference datasets. Of these, matching allele-frequency data for the predefined reported allele were available in at least one of the two gnomAD datasets for 103 variants, whereas for 9 variants the VEP-based query did not return a matching gnomAD frequency for that allele. Frequencies were reported for the same predefined reported allele across all groups, and differences between the study groups were assessed using 95% confidence intervals, Fisher’s exact tests, and false discovery rate correction. Genotype counts and the proportions of individuals carrying at least one copy of the reported allele were also summarized for all selected variants. Several pharmacogenomic variants showed population-specific frequency patterns, including NUDT15 rs116855232, SLCO1B1 rs4149056, VKORC1 rs9934438, and UGT1A1 rs10929302. Comparison with gnomAD showed that the observed frequencies were variant-specific and could not be consistently approximated by a single broad genetic ancestry group. Reference-based population structure analysis provided additional ancestry context and supported separate reporting by population group. The study did not evaluate clinical outcomes or make individual prescribing recommendations, and the small Uzbek sample size limits the precision of frequency estimates for this group, particularly for rare variants. Overall, this study provides a clinically prioritized pharmacogenomic frequency resource for underrepresented population groups in Kazakhstan and supports broader Central Asian representation in pharmacogenomic implementation research. Full article

Agricultural decision-making often depends on timely, evidence-based advice, yet advisory systems for underserved farming communities must also be cost-effective enough to operate continuously. Retrieval-augmented generation (RAG) offers a promising way to provide grounded agricultural recommendations, yet it remains unclear which commonly proposed techniques consistently improve answer quality in real-world deployments. To investigate this question, we developed a bilingual English–Spanish RAG advisory system for Arkansas rice, soybean, and poultry production and used it as a production-oriented testbed under a strict zero-marginal-cost deployment constraint. Through controlled paired experiments evaluated by an independent LLM judge, we compared retrieval strategies, prompt engineering techniques, language models, and evaluation protocols. Several widely used retrieval techniques, including BM25, HyDE, query rewriting, reranking, and token-based chunking, did not improve end-to-end answer correctness over a dense retrieval baseline. In contrast, structured prompting with exemplars consistently improved correctness, while alternative prompt formulations and the freely available language models evaluated offered no measurable advantage over the incumbent 70B model. We also show that conventional single-reference evaluation can substantially underestimate answer correctness in redundant knowledge bases and that independent LLM judging provides a more conservative assessment than self-judging. Together, these results provide practical guidance on which interventions improve answer quality in production-oriented RAG systems and introduce a human-validated multi-reference evaluation protocol for more reliable assessment of answer correctness. Full article

Selecting an appropriate visualization technique for multidimensional and multivariate financial and accounting (F&A) data remains a complex, user-dependent task. The TaxUI&BV4FADA taxonomy previously organized this problem along four dimensions—visualization techniques, user intentions and analytical goals, interaction possibilities, and user groups—but as a human-readable structure, it could not be queried, validated, or integrated into semantic decision-support pipelines. This paper presents an ontological framework that extends TaxUI&BV4FADA into a machine-readable OWL DL artifact authored in WebProtégé, with OWL used for semantic structuring and SPARQL used for score-based recommendation retrieval. The framework formalizes the four taxonomy dimensions and adds a decision-support layer and an evaluation layer. An explicit F&A semantic mapping is provided, and two contrasting worked scenarios—a financial analyst testing a gross-margin hypothesis and a CFO seeking a quarterly overview—show that the framework discriminates between F&A roles and analytical tasks. The evaluation demonstrates logical consistency, competency-question satisfaction, and internal consistency of the populated recommendation matrix against taxonomy-derived expectations, rather than independent empirical recommendation accuracy. This constitutes an internal, artifact-centered validation rather than an external empirical study with end users, and a protocol for future empirical validation with financial and accounting professionals is outlined. The framework provides a domain-oriented semantic and matrix-based decision-support foundation on which executable F&A visualization recommenders can be built. Full article

Electric Vehicle Charging Systems (EVCSs) have become a critical component of the global transition toward sustainable and intelligent transportation. However, their tight integration with heterogeneous cyber–physical, vehicular, and cloud-based infrastructures exposes them to an expanding attack surface, including data poisoning, malware injection, denial-of-service, and man-in-the-middle (MITM) attacks. Existing security solutions largely rely on isolated detection mechanisms and lack a unified semantic representation of EVCS assets, attack propagation paths, and mitigation dependencies, limiting their effectiveness in complex and evolving threat scenarios. To address these challenges, this paper proposes EVCS-SecOnt, an ontology-driven cybersecurity framework for modeling, reasoning, and mitigating security threats in EVCS infrastructures. The proposed ontology formalizes relationships across four core modules, namely Attack Surface, Attack Classification, Protection Mechanisms, and Risk and Mitigation, enabling holistic threat representation and TARA-based risk assessment. EVCS-SecOnt incorporates standard semantic namespaces (em:, seas:, uiote:, sch:, and time:) to ensure interoperability and is instantiated using the CICEVSE2024 dataset to support observation-level security reasoning. A unified SPARQL-based analytical workflow is employed to perform global ontology validation, attack–risk–severity correlation, mitigation prioritization, and observation-level inference using statistical feature vectors. Experimental results demonstrate that the ontology captures multiple attack classes, risk levels, severity categories, and mitigation strategies, enabling automated identification of critical attack scenarios and context-aware defense recommendations. The validation demonstrates logical consistency, semantic traceability, and query-based coverage of the ontology across attack classes, risk levels, severity categories, and mitigation strategies. EVCS-SecOnt enhances the interpretability, reusability, and explainability of EVCS cybersecurity management by bridging operational data with semantic intelligence. The proposed framework supports adaptive protection, risk-aware decision-making, and ontology-driven security analytics, providing a semantic foundation for next-generation e-mobility and smart charging infrastructures. Full article

Large language models are increasingly embedded in e-commerce assistants that recommend links to shopping destinations. When the candidate set contains adversarially crafted phishing URLs, a fluent model can recommend a malicious link with the same confidence it recommends a legitimate one, turning a helpful assistant into a delivery channel for fraud. Rather than treating phishing defense as per-link URL classification, this work frames the problem as recommendation-level exposure minimization: the output set itself must be secured, and a recommendation is counted as useful and safe only when it surfaces a benign destination and exposes no phishing URL. We first organize phishing URL constructions into four families spanning brand padding, typosquatting, homograph substitution, and subdomain impersonation, and use them to build candidate pools that mix benign links with plausible distractors. Evaluating four widely used open models under prompt-only defenses reveals a persistent gap: the strongest prompt baseline reaches only a 47.4 percent four-model average useful-safe rate, and weaker models fall below 20 percent even after careful prompting. We then present SAFER, a Security-Adaptive Filtering framework for Exposure-Minimized URL Recommendation. SAFER separates contextual selection from safety admission by coupling a deterministic lexical and structural pre-filter, an evidence-augmented single reasoning pass, and a deny-by-default post-verification stage with a deterministic fallback. The same deterministic URL evidence is injected before generation to condition the model’s reasoning and reused after generation to constrain which model-selected URLs may reach the user. SAFER issues exactly one model call per query, matching the prompt baselines, so its gains come from structure rather than additional inference. Across the four models SAFER raises the average useful-safe rate to 87.0 percent, a 39.6-point improvement over the best prompt baseline, and the deny-by-default stage yields a positive net gain for every model, largest where the model is weakest. Shrinking the deterministic layer’s brand coverage in a held-out analysis degrades the pipeline gracefully rather than collapsing it, indicating that within the range we tested its robustness does not rest solely on memorizing a fixed brand list. Additional robustness experiments confirm that SAFER transfers to real phishing URLs from the OpenPhish feed, generalizes to unseen attack families, maintains zero exposure on hard benign negatives, outperforms supervised URL classifiers as an admission gate, and maintains exposure minimization under realistic pool structures within the evaluated threat model. Full article

To address the problems of complex process knowledge sources, heterogeneous representations, dispersed semantic associations, and limited reusability in the domain of machining distortion of thin-walled parts, this study proposes a knowledge graph construction method for the workpiece machining distortion domain, together with an intelligent decision-making framework driven by the collaboration of knowledge graphs and large language models. First, a domain ontology model is established around core concepts, including workpiece objects, deformation-driving factors, analytical resources, analytical methods, and optimization knowledge, thereby providing a unified semantic foundation for domain knowledge organization. Second, considering the characteristics of domain texts, such as dense technical terminology, ambiguous entity boundaries, and complex relation expressions, a dual-channel knowledge extraction method integrating BERT-BiLSTM-CRF and Universal Information Extraction (UIE) is developed to achieve high-precision extraction of entities and relations from unstructured texts. Knowledge fusion is further carried out through cross-validation, entity disambiguation, coreference resolution, and semantic alignment, and the extracted knowledge is ultimately stored and organized in Neo4j. Furthermore, an intelligent decision-making framework based on the collaboration of knowledge graphs and large language models is constructed. In this framework, a LoRA-tuned Qwen model is employed for user intent recognition and key information extraction, RapidFuzz WRatio is adopted for similar-node retrieval, and local subgraph construction, Label Propagation-based community detection, Betweenness Centrality-based key-node analysis, and evidence fusion are integrated to support process recommendation and intelligent question answering. Based on the proposed framework, an intelligent decision-making system is further developed for process recommendation and intelligent question answering in machining distortion scenarios. Experimental results show that the proposed dual-channel knowledge extraction model achieves an F1-score of 0.88, demonstrating its effectiveness in knowledge acquisition for the machining distortion domain. The constructed knowledge graph contains 4639 entities and 5822 relations, enabling a systematic representation of machining distortion knowledge. Case studies further demonstrate that the proposed method can generate interpretable recommendation results under complex process constraints in real industrial query scenarios. Overall, the proposed approach provides a feasible pathway for the structured organization, intelligent retrieval, and decision support of workpiece machining distortion knowledge. Full article

The increasing digitization of healthcare records and growing reliance on interconnected systems have increased hospitals’ exposure to cyber threats, including ransomware, brute-force attacks, and Structured Query Language (SQL) injection. Traditional cybersecurity approaches alone are often insufficient to address these threats, prompting growing interest in artificial intelligence (AI)-based decision-support tools. This paper evaluates the potential of ChatGPT for hospital cybersecurity and incident response while introducing a structured qualitative framework for evaluating Large Language Model (LLM)-generated cybersecurity recommendations in healthcare. Through three progressively designed experiments and a ransomware case study, we evaluate ChatGPT’s role in developing a hospital cybersecurity plan, detecting brute-force login attempts, responding to an SQL injection attack, and managing a ransomware incident. Responses are assessed using five evaluation dimensions: specificity, completeness, technical correctness, feasibility, and alignment with the National Institute of Standards and Technology (NIST) Cybersecurity Framework (CSF), including both explicit mapping and function coverage. The results demonstrate that ChatGPT provides structured, context-aware guidance that aligns well with NIST CSF 2.0 and addresses governance and third-party risks. However, the recommendations also exhibit limitations, including limited operational depth, assumptions about technology and regulatory environments, lack of prioritization for resource-constrained settings, and limited consideration of implementation costs. Overall, the proposed evaluation framework provides a systematic approach for assessing LLM-generated cybersecurity guidance, while the findings indicate that ChatGPT can serve as a valuable decision-support tool that should complement, rather than replace, qualified cybersecurity professionals. Full article

Background/Objectives: Penile cancer carries a disproportionate burden of stigma and delayed presentation, with affected men increasingly turning to artificial intelligence (AI) chatbots as an anonymous information source. This study evaluated the quality, readability, understandability, actionability and clinical accuracy of AI chatbot responses to standardised penile cancer patient queries across six publicly available platforms. Methods: Fourteen standardised questions were submitted to ChatGPT-4o, Gemini, Perplexity, Microsoft Copilot, Claude, and DeepSeek, generating 84 responses. The responses were evaluated using DISCERN (information quality, scored 16–80), Patient Education Materials Assessment Tool for Printable Materials (PEMAT-P) (understandability and actionability, scored 0–100%), and Flesch–Kincaid grade level (reading complexity, recommended threshold ≤Grade 8). Between-platform and between-domain comparisons used the Kruskal–Wallis test with Dunn’s post hoc analysis. Clinical accuracy was assessed against the 2026 European Association of Urology-American Society of Clinical Oncology (EAU-ASCO) Collaborative Guidelines on Penile Cancer using a three-point ordinal scale across 10 guideline-scorable questions (maximum 20 points per platform); the four questions not addressed by the guidelines were scored separately for factual accuracy against authoritative external evidence. Results: No significant between-platform differences were identified for DISCERN (p = 0.796) or PEMAT-P understandability (p = 0.147). All platforms exceeded the 70% understandability adequacy threshold. Perplexity and Copilot demonstrated significantly higher actionability than all other platforms (100% vs. 75%, p < 0.001). No platforms achieved the recommended Grade 8 reading threshold, with Claude generating significantly more complex responses than ChatGPT-4o and Perplexity (grade 11.4 vs. 8.65 and 8.60, p < 0.05). Significant variation was identified across clinical domains for DISCERN (p < 0.001), Flesch–Kincaid grade level (p < 0.001), and word count (p < 0.001), with treatment-related questions achieving the highest information quality but also generating the most complex responses. Clinical accuracy scores ranged from 13/20 (ChatGPT-4o) to 16/20 (Copilot). Two critical errors were identified: Claude and DeepSeek both recommended bleomycin-containing chemotherapy regimens, directly contradicting the EAU-ASCO Strong recommendation against bleomycin due to pulmonary toxicity risk. Responses to the four survivorship and quality-of-life questions were factually accurate against external evidence in 23 of 24 cases. Conclusions: AI chatbot responses to penile cancer patient queries are broadly understandable but consistently fail to meet recommended readability thresholds and provide limited actionable guidance. Two platforms recommended bleomycin-containing regimens against an EAU-ASCO Strong recommendation; no platforms achieved full guideline concordance. Urologists should counsel patients on the limitations of AI chatbots as a health information source. Full article

Background/Objectives: Accurate arterial tree reconstruction from computed tomography angiography (CTA) is essential for vascular diagnosis, surgical planning, and hemodynamic modelling. A persistent and underappreciated problem is vessel discontinuity: thin distal branches appear as disconnected fragments rather than continuous structures, caused by bifurcations, image noise, contrast variation, arterial plaque, motion artifacts, and partial volume effects. This scoping review aimed to systematically characterize computational approaches addressing vessel breakage in CTA segmentation and identify methodological gaps warranting further investigation. Methods: This scoping review was conducted in accordance with PRISMA-ScR guidelines. Google Scholar and PubMed were queried for studies published between March 2000 and March 2026. Eligible studies included peer-reviewed journal articles and conference proceedings presenting original methodological contributions to three-dimensional vascular segmentation from CTA. Results: Three generations of computational solutions were identified: classical geometric and geodesic methods, deep learning approaches with topology-aware training, and hybrid post-processing frameworks. Topology-sensitive metrics (clDice, Topology Sensitivity) were identified as preferred metrics to better capture clinical utility than standard voxel-based metrics such as the Dice coefficient. Conclusions: Vessel discontinuity remains a clinically relevant and challenge in vascular CTA segmentation. Hybrid post-processing frameworks combining deep learning with geodesic connectivity restoration represent the current state of the art. Standardized adoption of topology-aware evaluation metrics is recommended to better reflect clinical utility in future studies. Full article

Background: Artificial intelligence and large language models (LLMs) are emerging as transformative technologies in medicine. However, their ability to develop physical exercise rehabilitation programs and provide insights into musculoskeletal (MSK) disorders remains underexplored. This study aimed to evaluate the quality and readability of LLM-generated responses to consultation questions addressing various stages of the clinical process encountered by patients with MSK disorders. Methods: This study recruited 50 patients with musculoskeletal disorders and extracted disease-related frequently asked questions from Google search. We developed three clinical scenario-based question types simulating real consultations, which were processed by four LLMs (GPT-3.5-turbo, GPT-4-turbo, GPT-4o, and Claude-3-haiku-20240307). Response quality was assessed by orthopedic specialists and therapists using the DISCERN instrument, and GPT-4o-assisted evaluation was used to extend the assessment after validation with expert ratings. Readability was systematically assessed via six validated indices. Results: Among the 1476 LLM-generated responses, the generated rehabilitation programs demonstrated consistent adherence to the specified query requirements. The DISCERN scores ranged from 26 (poor) to 68 (excellent), with a mean score of 55.60 ± 8.40. The intraclass correlation coefficient (0.68) indicated moderate interrater agreement, and Cronbach’s α showed good internal consistency. The readability scores across six indices indicated that most responses exceeded the recommended reading levels (p < 0.05). Conclusions: LLMs generated moderate-to-high-quality PE rehabilitation recommendations for patients with MSK disorders across simulated consultation scenarios. However, limited supporting materials and suboptimal readability may restrict their effectiveness. With physician oversight, improved readability, and enhanced supplementary resources, LLMs demonstrate considerable potential as supportive tools in orthopedics. Full article

In today’s increasingly complicated network environments, effective cybersecurity analytics necessitate scalable data processing systems that can handle massive amounts of diverse traffic data. This article compares relational, graph-based, and streaming data systems for cybersecurity analytics using the CICIDS2017 dataset. We specifically compare a columnar cloud data warehouse (Amazon Redshift) with a graph database (Neo4j) using example analytical queries to investigate trade-offs in query expressiveness, performance, and data modeling flexibility. In addition, we evaluate a real-time data intake pipeline built on Apache Kafka and Apache Cassandra to investigate ingestion throughput and low-latency storage features under simulated streaming workloads. The systems are examined independently to highlight their strengths and weaknesses in batch analytics, relationship-centric analysis, and real-time monitoring. The findings offer practical insights into how alternative data models and processing paradigms impact cybersecurity analytical tasks, as well as recommendations for selecting optimal data systems for network traffic analytics and intrusion detection use cases. Full article

Multimodal knowledge graph completion (MMKGC) aims to predict missing entities by exploiting structural, visual, and textual information and is important for semantic retrieval, recommendation, and intelligent question answering. Existing relation-aware methods usually use relational context to adjust modality weights or fuse multimodal scores but rarely exploit the historical answer distribution of each directed relation as explicit ranking evidence. To address this limitation, we propose RAPC, a Relation Answer Space Prototype Calibration framework. For each query, RAPC obtains multimodal prior scores from structural, visual, textual, and image–text cross-modal branches, retrieves the top-k query-relevant anchors from the historical answer space of the corresponding directed relation, and aggregates them into prototype evidence. This evidence is injected through a prior-preserving selective calibration mechanism, while relation-aware hard negative training improves discrimination between true answers and similar false candidates. Experiments on DB15K, MKG-W, and MKG-Y show that RAPC achieves clear gains on DB15K and MKG-W and obtains the best MRR and Hits@1 on MKG-Y. Compared with the best external results, RAPC improves MRR and Hits@1 by 2.64 and 3.57 percentage points on DB15K and by 3.02 and 3.72 percentage points on MKG-W. These results show that relation-level historical answer distributions provide useful explicit evidence for top-ranked entity prediction. Full article

Building Information Modeling (BIM) usually represents a building as a static snapshot of the model’s state. Dynamic extensions, such as Internet of Things(IoT)-enabled sensing or immersive visualization, already exist, but the underlying data model remains state-based. The Industry Foundation Classes (IFC) standard does not define a formal mechanism that would link the same physical element across successive phases of a building’s life cycle. Design, construction, and operation are recorded in separate IFC files, and the same element is assigned different Globally Unique Identifiers (GUIDs) in each. The result is fragmentation of the element’s identity, loss of the history of property changes, and the inability to formulate cross-phase queries. This paper proposes the BIM-Phase ontology based on the fundamental Descriptive Ontology for Linguistic and Cognitive Engineering (DOLCE) ontology, which solves this problem by introducing a distinction between a building element as an endurant and its life cycle phases as perdurants. The ontology comprises nine classes, six object relations, and six axioms expressed in Web Ontology Language 2 Description Logic (OWL 2 DL). Phase properties and relations are represented using a reification pattern, which maintains full compatibility with the expressiveness of OWL 2 DL. The ontology was validated using an example of a single-family residential building developed in Autodesk Revit. Three structural elements (external wall, floor slab, and column) were tracked across three phases of the life cycle. Eight competency questions covering scalar, constitutional, and mereological changes were defined and mapped to ontology constructs, confirming that the BIM-Phase enables the recording of changes and the formulation of cross-phase queries that are impossible in classic IFC. All eight questions were answered correctly on the published knowledge graph, and the HermiT reasoner confirmed the logical consistency of the model. The findings show that preserving element identity across phases requires only a minimal ontological layer on top of existing standards. We recommend introducing persistent, phase-independent identifiers of building elements alongside IFC GUIDs, as this single change enables full lifecycle change tracking. Full article