Search Results (607)

Timely and accurate interpretation of Notices to Air Missions (NOTAMs) is essential for effective aeronautical information management (AIM). Although NOTAMs follow a standardized format, their abbreviated textual content and heterogeneous operational semantics make automated processing difficult. Existing methods often address type classification, information extraction, and semantic representation independently, leaving a gap between textual interpretation and ontology-grounded information management. To address this gap, this paper presents NOTE, an AI-driven ontology-grounded type-aware extraction framework for NOTAMs. NOTE is supported by a dynamic semantic ontology that combines a shared semantic module with seven type-specific sub-ontologies. Guided by the predicted semantic type, the framework activates the corresponding ontology-aligned schema contract, extracts and validates the relevant information, and materializes the result as Resource Description Framework (RDF) triples. Experiments on 25,341 operational NOTAMs and a manually curated ontology-mapping set produced a macro-$F_1$ of 0.9424 for type-aware routing and an $F_1$ of 0.9053 for structured extraction, with a schema conformance rate of 0.9976. The dynamic ontology achieved slot- and relation-level $F_1$ scores of 0.9472 and 0.9835, respectively. Compared with a scale-matched static ontology, type-specific activation improved relation mapping by 13.06 percentage points while producing semantic richness close to the human-annotated reference. These findings indicate that NOTE provides an effective connection between abbreviated NOTAM text and validated ontology-grounded representations for AIM. Full article

Utah Lake and Great Salt Lake share a watershed yet face distinct pressures: eutrophication, harmful algal blooms, hydrologic decline, and exposed-playa dust hazards, but their combined literature has never been systematically characterized. I analyzed 1383 peer-reviewed records using bibliographic coupling, Louvain community detection, latent Dirichlet allocation, and large language model-assisted annotation. The coupling network (1012 records, 7536 edges) split into 379 communities ($Q=0.541$), of which 342 were singletons and only 14 reached 10 or more papers, together holding 59% of coupled records. Of 10 main-text clusters, 6 are anchored in a lake and 4 in a topic rather than a system; just 1 is Utah Lake dominant. Great Salt Lake research spans broader disciplinary communities (brine-shrimp ecology, mercury cycling, dust and paleoclimate); Utah Lake research is more applied (native-species management, eutrophication, harmful algal blooms). Recent Utah Lake studies report no lake-wide chlorophyll-a trend across 1068 Landsat scenes (1984–2021) and dissolved phosphorus near ∼0.02–0.04 mg L⁻¹ over ∼50 years despite ∼300% population growth, implying internal sediment cycling dominates; recent Great Salt Lake work centers on hydrologic decline and playa dust hazards. The synthesis identifies five cross-cutting gaps, including dust-emission-to-exposure assessment, Utah Lake nutrient modeling, and predictive water management, and yields a reproducible, transferable framework. Full article

Financial contagion—the propagation of shocks across markets, sectors, and time—remains one of the central questions in empirical finance, and yet most computational approaches model it at the wrong granularity. Existing work captures contagion at the asset level by measuring realised return or volatility comovements between specific securities, which conflates a structural pattern with its surface manifestation. We argue that the genuine causal regularity in contagion is between event types (rate decisions, default announcements, regulatory actions) rather than between individual assets. This paper introduces TECG, a framework that learns time-stamped causal-temporal association edges—predictive dependencies rather than identified causal effects—between abstract financial event types from news text and uses them to forecast multi-step contagion cascades. Methodologically, TECG can be read as a neural impulse response function (IRF) for event sequences: where the canonical structural-VAR IRF traces the dynamic response of one continuous variable to a shock in another, TECG traces the conditional intensity with which one event type fires after a shock to another event type, at a learned and possibly multi-modal lag. The framework couples LLM-based event extraction with a multivariate neural Hawkes process whose intensity functions are parameterised by a temporal graph neural network. The resulting graph has interpretable, time-stamped edges of the form “event-type A triggers event-type B with lag distribution L and conditional intensity κ.” We evaluated TECG on an aggregated corpus of approximately 482,000 financial news items spanning 2007–2023 and report three findings. First, the learned edges recover associations consistent with known causal relationships in finance—central-bank announcements preceding sector-level earnings revisions, default events at one institution preceding due-diligence events at competitors—without supervision on these relationships. Second, edges learned from data up to 2018 transfer to held-out cascades from 2020 (COVID equity crash) and 2023 (US regional banking stress) with substantially better fidelity than baselines that ignore temporal structure or operate at the asset level. Third, the framework supports generalised impulse response analysis: given a hypothetical seed event, it produces a distribution over downstream event chains that an analyst can interrogate, time-stamp, and entity-resolve. We are explicit that the empirical validation is observational and that the causal interpretation rests on assumptions we discuss at length. All headline comparisons are supported by paired-bootstrap significance tests; the extraction front end is independently evaluated on a manually annotated benchmark, with an explicit error-propagation analysis; and additional comparisons against recent temporal point-process and temporal-graph baselines are reported. The framework’s principal limitation is that it does not separately identify endogenous fire-sale or leverage-spiral dynamics; we point to where these gaps could be closed. Full article

Pixel-level annotation remains a major bottleneck for semantic segmentation, motivating methods that synthesize image–label pairs directly from generative models. Prior synthetic dataset generators typically obtain pseudo-labels from cross-attention maps or learned decoders over generative features; however, recent text-to-image (T2I) models increasingly use multimodal diffusion transformers (MM-DiTs), where concept localization is no longer exposed through a single cross-attention pathway but instead distributed across many layers and attention heads. Existing MM-DiT localization methods address this by aggregating saliency across heads, but we observe that this averaging can dilute clean target localizers due to attention head heterogeneity. We introduce HEADHUNTER, a training-free segmentation framework that uses aggregate concept saliency as a self-guided proxy to select the single attention head that best localizes a queried textual concept, yielding cleaner segmentation masks. We then use HEADHUNTER to turn target classes into training data automatically: a large language model (LLM) diversifies prompts, an MM-DiT generates images, HEADHUNTER produces pseudo-labels, and a vision language model (VLM) verifies each image–mask pair before acceptance. HEADHUNTER achieves strong zero-shot segmentation performance (81.2 mIoU on PASCAL VOC2012 and 73.2 mIoU on ImageNet-Segmentation), outperforming head aggregation and other interpretability methods. Using only our generated image–label pairs, we train segmentation models which reach 64.9 mIoU on VOC2012 validation, matching or outperforming comparable synthetic dataset generators and showing that the proposed pipeline produces effective dense labels without human intervention. Full article

Preparing training data for domain-specific medical Named Entity Recognition (NER) involves a trade-off between annotation quality, expert effort, and data privacy: manual annotation is costly, whereas cloud-based Large Language Models (LLMs) raise concerns about the control of sensitive clinical text. This article introduces NERFlow, a workflow-driven subsystem of the FIT4NER project whose contribution is an abstraction layer that makes rule-based, model-based, and LLM-based annotation interchangeable and comparable within one configurable workflow environment. Open-source LLMs are integrated as exchangeable annotation services, deployable locally or in cloud-agnostic infrastructures via Kubernetes, and embedded into the KM-EP knowledge management system. NERFlow was evaluated qualitatively, through a cognitive walkthrough, an IEEE 1028 technical review, and a user-centered survey with 18 participants, and quantitatively on the CRAFT corpus with seven open-source and hosted LLMs run through an identical pipeline. The results support its use as LLM-assisted pre-annotation with expert correction, with locally deployable open-source models as the more reliable basis for reproducible operation. Full article

Automatic extraction of genealogical information from historical archival-genealogical documents in Uzbek is an understudied problem for low-resource languages. Multi-layer NLP benchmarks are not sufficient to automatically identify individuals, family relationships, dates, place names, and archival identifiers in such texts. Also, the same people are mentioned in various forms: full name, pronoun (18.8%), initial, surname-name order, indirect expression (9.4%), and title. Existing NER and relation extraction corpora are mainly focused on high-resource languages or general domain texts and do not sufficiently cover the FAMILY_ROLE signals, historical spelling variants, and fond–opis–delos identifiers specific to Uzbek archival-genealogical texts. Proposed resource: We present the ArchiveGene Corpus, a controlled, fully synthetic, and reproducible five-layer resource consisting of 1000 Uzbek archival-genealogical-style documents, divided into 700 training, 150 validation, and 150 test documents. The corpus contains 8366 named entities, 10,625 person mentions, 2000 coreference chains, and 1000 genealogical relation triples. The dataset was generated using a deterministic template-based pipeline and a lexicon of Uzbek names, and is fully reproducible. Inter-annotator agreement values were 0.847 for NER, 0.793 for coreference, and 0.821 for RE, according to Cohen’s κ. Comparative results are presented with four baseline models (rule-based, BiLSTM-CRF, mBERT, and XLM-RoBERTa). The dataset is openly hosted on the Zenodo platform under the CC BY 4.0 license; concept DOI: 10.5281/zenodo.20670360, v1.1.1 version DOI: 10.5281/zenodo. 21429998. Scientific significance: To the best of our knowledge, ArchiveGene is among the first openly released, controlled synthetic resources for Uzbek that integrates named-entity recognition, person-mention detection, coreference resolution, genealogical relation extraction, and final tuple generation within a single annotation framework. The baseline analysis provides three main conclusions: (1) on the clean synthetic test set, the transformer models already reach 100.00 Micro-F1 for NER and 100.00 Macro-F1 for coreference-aware relation extraction, so coreference aggregation adds little on synthetic data (+2.25 for mBERT and +0.04 for XLM-RoBERTa) but its contribution is expected to grow on real archival text; (2) the rule-based and heuristic baselines lag far behind (Macro-F1 50.28 and 70.73) and fail entirely on spouse_of, showing the limits of lexical rules; and (3) a zero-shot evaluation on a real-document pilot reduces NER Micro-F1 from 100.00 to 22.17, indicating that the synthetic corpus is trivially learnable and that real-archival validation is essential. Full article

Background and Objectives: Cardiology departments generate large volumes of unstructured free-text reports that impose substantial manual review burdens on clinicians; at Hospital de Santa Maria—Portugal’s largest public hospital—manual review of 12,651 reports took approximately seven minutes per report, representing over 1475 h of avoidable administrative work. This study presents Text2FHIRwallet, a health professional digital wallet that automates extraction and structuring of clinical entities from unstructured Portuguese cardiology reports using fine-tuned Named Entity Recognition (NER) models and maps the results to Fast Healthcare Interoperability Resources (FHIR) R4 patient summaries. Materials and Methods: Following the Design Science Research Methodology (DSRM) and CRISP-DM, we fine-tuned four transformer-based models—BERTimbau Base, BERTimbau Large, Albertina PT-PT, and MediAlbertina—on 305 manually annotated cardiology reports (77,309 tokens; κ = 0.85 inter-annotator agreement) covering eight clinical entity types, drawn from a corpus of 12,651 anonymised documents. Entities were mapped to FHIR R4 resources and delivered through a secure, role-based mobile wallet (React Native). Evaluation comprised token-level NER benchmarking with bootstrapped confidence intervals and McNemar’s testing, FHIR mapping accuracy assessment on 100 manually reviewed reports, processing-efficiency measurement, and a usability pilot with 10 cardiologists (SUS, NPS). Results: MediAlbertina achieved the highest NER performance (macro F1 = 0.985, 95% CI: 0.979–0.990), significantly outperforming all baseline models (p < 0.01, McNemar’s test) and comparing favourably with—though not directly comparable to, given differing languages and datasets—published benchmarks such as GPT-4 (F1 = 0.962 in ophthalmology NER) and fine-tuned BERT models for lung cancer NER (F1 ≈ 0.85–0.90). FHIR mapping accuracy was 98% on 100 independently reviewed reports. Report processing time was reduced from approximately seven minutes to 15–30 s (93–96% reduction), with peak batch-inference throughput of up to 1000 reports/h under parallelised GPU load (observed end-to-end throughput in pilot deployment was approximately 250 reports/h). The pilot usability evaluation yielded a SUS score of 87 (excellent) and an NPS of 80. Conclusions: Text2FHIRwallet demonstrates that domain-specific fine-tuning of a Portuguese-language pretrained language model achieves near-ceiling clinical NER accuracy, enabling scalable, interoperable, and privacy-compliant patient summary generation from unstructured cardiology text, offering an end-to-end pathway for integrating AI-driven NLP into clinical workflows and FHIR-based health information ecosystems, with implications for administrative efficiency, care coordination, and clinical research in non-English-language settings. Full article

Weakly supervised video anomaly detection (WS-VAD) localizes anomalous events in untrimmed videos using only video-level annotations. While CLIP-based methods have advanced this task through vision–language alignment, widely adopted approaches construct text prototypes from short category-name prompts of at most five words, leaving the CLIP text encoder not fully exploited. We propose SETAS-VAD, which addresses this gap through a Category Semantic Alignment (CSA) loss function: for each anomaly category, a large language model generates multi-sentence descriptions covering complementary semantic aspects, encoded once offline into frozen prototype vectors. An InfoNCE contrastive objective pulls attention-weighted anomaly features toward ground-truth category prototypes at zero additional inference overhead (prototype generation and encoding are performed once offline as a preprocessing step, not at test time). Under fully reproducible conditions on UCF-Crime and XD-Violence, SETAS-VAD achieves state-of-the-art temporal localization (30.45% mAP on XD-Violence, 12.16% on UCF-Crime), with per-threshold gains increasing at stricter IoU values, indicating improved boundary precision rather than coarse detection sensitivity. Full article

In live-streaming e-commerce systems, massive volumes of user-generated danmaku constitute a critical yet underutilized source of business intelligence. However, transforming such unstructured, noisy, and highly context-dependent textual data into structured and actionable knowledge remains a fundamental challenge for enterprise information systems. To address this issue, this study proposes TaSC-LLM, an LLM-enabled topic recognition method for constructing interpretable topic measurements from unstructured user-generated content. The proposed framework integrates topic taxonomy construction and zero-shot classification into a unified semantic reasoning pipeline. Unlike conventional topic modeling or supervised classification approaches, TaSC-LLM leverages chain-of-thought reasoning, multi-stage taxonomy induction, sliding window context modeling, and self-consistency verification to eliminate reliance on predefined label spaces and annotated training data. This design allows the system to dynamically construct and update topic taxonomies while ensuring interpretability, robustness, and cross-scenario adaptability. Empirical evaluation on three large-scale live-streaming e-commerce danmaku datasets shows that TaSC-LLM achieves strong taxonomy coverage, classification accuracy, and agreement with expert annotations. The findings suggest that LLM-based reasoning can help convert unstructured user-generated text into interpretable topic measures for downstream empirical and managerial analysis. While the present evaluation is conducted offline, TaSC-LLM provides a methodological foundation for future business applications that can be further examined under multi-session, multi-platform, and deployment-oriented conditions. Full article

Large language models (LLMs) exhibit strong general conversational capabilities, yet their deployment in domain-specific service environments such as hospitality remains limited by the absence of emotionally grounded datasets and validated end-to-end system architectures. This paper presents HOSPIT-LLM, an EU-funded euROBIN Technology Exchange Program pilot, as a complete, integrated pilot pipeline for human-centered conversational AI in hotel reception scenarios. We deploy a multimodal hotel-terminal assistant in a real hotel reception, capturing synchronized dual-camera video and audio to collect authentic guest–staff interactions. Speech is transcribed using Whisper, and emotion is extracted from the corresponding video segments via DeepFace, producing 582 real Greek guest–receptionist exchange examples. The resulting data are classified into eight Standard Operating Procedure (SOP) categories. To address data scarcity, we augment the corpus with 1269 synthetic dialogues generated by eight diverse LLMs through the OpenRouter API, yielding a total of 1851 dialogue records with explicit emotion-token annotation. We fine-tune Qwen3.5-35B-A3B using Low-Rank Adaptation (LoRA) through a two-stage process: supervised fine-tuning (SFT) on an 888-example conversation pool and Simple Preference Optimization (SimPO) on a 1899-pair preference pool, each split 80/10/10 into training, validation, and test. The resulting model is integrated into an interactive hotel-terminal system combining YOLO-based person detection, face-recognition-driven guest personalization, Kokoro neural text-to-speech (TTS), and a multi-service orchestration layer connected to the hotel Property Management System (PMS). Evaluation combines standard text metrics, emotion-aware scoring, and a large-model judge. The results indicate targeted improvements in the rule-based contextual emotion-policy match and staff-emotion policy compliance compared to the base model, while general response-quality gains remain more modest. In particular, the rule-based contextual policy-match score improves from 0.614 to 0.901, while forbidden staff-emotion outputs decrease from 0.142 to 0.018. The deployed pilot demonstrates the practical integration of a personalized, emotion-aware LLM assistant in an interactive hotel-terminal setting; end-to-end latency and fully hotel-side edge deployment were not evaluated and are left for future work. HOSPIT-LLM provides a reproducible framework for multimodal dataset creation, preference-based fine-tuning, and deployment of human-centered AI systems. A mobile robotic embodiment is planned as future work. Full article

Sonar object detection is constrained by scarce instance-level annotations, high acquisition costs, and long-tailed category distributions in real underwater environments. To address these limitations, this paper proposes a generative data augmentation framework based on the Stable Diffusion Model (SDM) for synthesizing sonar images together with target bounding boxes. The framework first fine-tunes SDM with an instance-level slice cropping strategy to strengthen the alignment between text prompts and local acoustic target structures. It then introduces a cross-modal sparse localization module (CMSL), which uses denoising features and text priors to infer 2D bounding boxes for generated samples. Synthetic long-tail samples are mixed with real URPC2022 data under a fixed-ratio saturation compensation strategy and evaluated through UTD-SCnet fine-tuning. The results show that instance-level cropping provides the best generation quality among the tested strategies (FID = 31.42, DR = 0.815, CCSCR = 0.075), and that a 60% synthetic-data injection ratio yields the best detection performance. These findings indicate that diffusion-based augmentation can provide a practical, semi-automated supplement for long-tail sonar detection, while excessive synthetic data may introduce domain-shift effects. Full article

Multi-expert large language model systems need to dynamically distribute user requests among multiple candidate experts to improve inference accuracy and system efficiency in complex task scenarios. To address the issues of existing routing methods relying on manual rules, unified discriminant boundaries, or large-scale annotated data, as well as high maintenance costs during expert expansion, this paper proposes an expert routing method, BGMM-LSE, based on positive sample semantic distribution modeling. The method constructs a positive sample set using only the requests historically successfully processed by each candidate expert, and maps the requests into dense semantic vectors through a pre-trained text feature extraction model; subsequently, a Bayesian Gaussian mixture model (BGMM) is independently trained for each expert to characterize the capability distribution of its successful requests in the semantic space. During online inference, the system encodes the target request into a feature vector, calculates the log-likelihood score combining the retained Gaussian component parameters and the smoothed covariance matrix of each expert, and uses Log-Sum-Exp to aggregate the probability contributions of different components. Finally, the request is routed to the expert model(s) with the highest score. Experiments were evaluated based on seven tasks: Math, GSM-Symbolic, HumanEval, Mbpp, MMLU, AIME1983–2025, and HellaSwag. The results show that this method achieves an average routing accuracy of 86.25%. The experimental results demonstrate that BGMM-LSE can provide stable request distribution capabilities for multi-expert large language model systems while maintaining interpretability and scalability. Full article

Breast cancer remains one of the leading causes of cancer-related mortality among women worldwide, where early and accurate diagnosis is critical for effective treatment. Although recent advances in deep learning have enabled automated radiology report generation from breast ultrasound images, most existing approaches function as black-box systems, limiting clinical trust and interpretability. This study proposes a trustworthy and explainable framework for automated breast ultrasound report generation that combines Vision-Language Modelling (VLM) with multi-level Explainable Artificial Intelligence (XAI). The proposed architecture integrates a Swin Transformer for visual feature extraction, BioBERT/ClinicalBERT for clinical text representation, and a GPT-2-based decoder for report generation through a dual cross-attention fusion mechanism. The framework is evaluated on benchmark breast ultrasound datasets paired with expert-annotated radiology reports using standard natural language generation metrics, including BLEU, ROUGE-L, METEOR, and CIDEr. Experimental results demonstrate that the multimodal architecture significantly improves report quality, clinical consistency, and semantic accuracy compared with conventional image-only and single-modal baselines. To address transparency and trustworthiness, the framework provides dual-level explanations through Grad-CAM visual heatmaps and LIME/SHAP-based token attribution analysis, enabling clinicians to understand both image regions and textual features influencing generated reports. Qualitative assessment further indicates strong alignment between model explanations and radiologist-identified diagnostic findings. Full article

Sentiment analysis for Greek social media texts remains challenging because of limited annotated resources, linguistic variation, and domain-dependent sentiment expression. This study presents a topic-aware, lexicon-guided framework for sentiment classification across five reference domains in Greek social media. Domain-specific sentiment lexicons are activated according to the relevant domain and transformed into a structured representation comprising a token-level multi-channel lexical matrix and aggregate lexical descriptors. A fusion convolutional neural network combines these complementary components to classify sentiment while retaining explicit lexical evidence for inspection. The evaluation follows a leakage-free protocol: lexicons are constructed exclusively from the sentiment inner-training subset, validation data are used for model selection, and a held-out test set is reserved for final evaluation. The proposed fusion CNN achieved the strongest held-out sentiment result among the evaluated models, with an Accuracy of 0.8029 and a Macro-F1 of 0.7883, exceeding TF–IDF + Linear SVM and fine-tuned GreekBERT baselines in the present experimental setting. Ablation results show that the token-level lexical matrix and global descriptors provide complementary information. For domain routing, GreekBERT late fusion achieved an Accuracy of 0.9162 and a Macro-F1 of 0.9116. When lexicon activation used predicted rather than reference domains, the end-to-end sentiment pipeline achieved a Macro-F1 of 0.7569. These findings indicate that explicit domain-specific lexical knowledge can support an interpretable sentiment representation while making the effects of lexical coverage and topic-routing uncertainty visible. Full article

Industrial maintenance and fault-diagnosis systems generate textual records, such as maintenance work orders, service requests, causal analyses, and troubleshooting solutions. These records contain domain-specific named entities that provide valuable knowledge for intelligent monitoring, fault diagnosis, maintenance decision support, and industrial knowledge graph construction. However, in practical industrial environments, maintenance records are strongly associated with specific equipment types, production processes, fault modes, and enterprise-specific terminology. As a result, entity schemas vary across systems, new entity types emerge with equipment updates, and high-quality annotation requires substantial domain expertise. These factors make it difficult to obtain sufficient labeled samples for each industrial entity type. Under such low-resource conditions, conventional supervised named entity recognition (NER) models tend to suffer from unstable entity boundary detection and biased entity representations. To address these challenges, this paper proposes a boundary-aware knowledge graph-guided adaptive prototype correction framework for few-shot NER in industrial maintenance information systems. The proposed framework first introduces a boundary-aware span detection mechanism to improve entity localization in noisy and irregular maintenance texts. A knowledge graph-guided adaptive prototype correction module is then designed to construct entity class prototypes from limited support examples, reducing prototype bias caused by sparse annotations. Experiments are conducted on two representative industrial datasets, MaintIE and CFDK, covering maintenance short texts and fault-diagnosis records. Experimental results show that the proposed framework achieves an average Micro-F1 improvement of 1.89 percentage points over the strongest compared baseline across 12 episodic settings on the two industrial datasets: three MaintIE coarse-grained settings, six MaintIE fine-grained settings, and three CFDK settings. The ablation and sensitivity analyses further indicate that boundary-aware span modeling and KG-guided prototype correction jointly contribute to low-resource entity classification. This study provides a data-efficient information extraction solution for AI-enabled industrial knowledge acquisition, fault diagnosis, and maintenance decision support. Full article