Search Results (278)

As urban renewal progresses, transforming old industrial plants into sports service complexes has become a prevalent strategy to revitalize urban stock space. To evaluate the actual performance of this spatial reproduction from the users’ perspective, this study analyzes 3702 consumer reviews from five typical Chinese sports service complexes using an interdisciplinary text-mining framework. We employ the LDA topic model to extract core dimensions of consumer concern and utilize the DistilBERT model for fine-grained, sentence-level sentiment computation. The results reveal that consumer attention spans five key topics: venue services, cultural business districts, sports training, industrial integration, and heritage utilization. Critically, the sentiment analysis uncovers a structural paradox: while topics associated with cultural and heritage utilization trigger highly positive emotions and strong place identity, core functional modules like sports training and venue services generate substantial negative feedback. This disparity highlights a profound friction between rigid historical industrial architectural structures and the flexible demands of modern servicescapes, reflecting a clear path dependency in spatial transformation. Based on these theoretical and empirical findings, we propose targeted optimization strategies, including the structural reconstruction of training services and the flexible upgrading of basic venue facilities, to transition these complexes from initial physical construction toward long-term, service-oriented operation. Full article

We compare six purposively selected term pairs using an English-language web corpus composed mainly of encyclopedic sources: engawa/veranda, tsuboniwa/impluvium, shōji/fusuma, tokonoma/chigaidana, ma/mu, and wabi/sabi. The pairs form three groups: cross-cultural comparisons, related Japanese architectural elements, and conceptual or aesthetic pairings. The study applies an audited corpus-based Bidirectional Encoder Representations from Transformers (BERT) analysis. Each pair is tested across predefined seed-based fields under FULL corpus conditions and CLEAN conditions that remove glossary-like and explicitly definitional sentences. In the tested corpus, the pairs do not behave as simple equivalents or oppositions. In cross-cultural comparisons, engawa is more strongly associated with indoor–outdoor boundary vocabulary, while narrowly specified boundary and transitional probes also favor engawa; veranda is more strongly linked to circulation and sheltered outdoor use. Tsuboniwa leans toward meanings of miniature gardens, whereas impluvium is strongly associated with water, basins, and Roman domestic systems. Within Japanese architectural vocabulary, fusuma dominates partition-related fields, whereas tokonoma and chigaidana remain mostly weak or non-separating. Among abstract terms, ma and mu show an asymmetric bidirectional pattern, with correction-supported separation more consistent for ma-directed than for mu-directed fields. Wabi and sabi show high contextual similarity in the selected English-language corpus, with correction-supported separation only for impermanence and melancholy. The field comparisons identify where the paired terms differ, where no difference was detected, and where one term was more strongly associated with several fields. The findings concern contextual usage in this corpus rather than definitive architectural, historical, or philosophical meanings of the terms. Full article

The proliferation of algorithmically generated malicious URLs presents a critical challenge for modern cybersecurity, requiring a shift from syntactic pattern-matching toward semantic understanding grounded in pre-trained transformer foundations. Building on our prior work on billion-scale semantic search and density-based campaign clustering, this paper presents a unified, deployable framework for real-time campaign discovery. The framework converts raw URL streams into dense Sentence-BERT embeddings and couples approximate nearest neighbor search with online density-based clustering, discovering emerging campaigns without prior knowledge of their number or shape. Our central finding is that the choice of semantic representation is decisive: a domain-focused embedding strategy yields near-perfect campaign separation, substantially outperforming full-URL representations. On live, in-the-wild threat feeds, the domain-focused representation recovers all 944 discovered campaigns at an Adjusted Rand Index of 0.990 and a mean campaign recall of 1.000, at 0.10 ms per URL. Under identical clustering, the full-URL representation reaches an Adjusted Rand Index of only 0.510 and recovers fewer than half of the campaigns. We add a SHAP-based explainability layer that attributes discovery decisions to interpretable structural patterns, and we expose the whole pipeline as an operational system with a REST interface for single- and batch-URL analysis. The encoder we employ is a discriminative representation model rather than a generative one; what we contribute is the representation-and-modeling discipline this setting demands, and the resulting guidance transfers to threat intelligence in an era of generatively produced attack campaigns. Full article

Large Language Models (LLMs) show promise for supporting formative assessment by generating feedback on students’ written mathematical reasoning. However, practical use in educational settings remains constrained by the need to manually curate representative few-shot exemplars for prompt construction. This study examines whether unsupervised semantic clustering can automate few-shot exemplar selection for LLM-generated formative feedback in middle-school mathematics. As a controlled methodological feasibility study, we generated and refined 100 exam-realistic constructed responses for a Grade 7 inequality task aligned with middle-school mathematics standards. Student responses were embedded using Sentence-BERT, projected into a lower-dimensional space using Uniform Manifold Approximation and Projection (UMAP), and clustered with Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) to identify dominant reasoning patterns and ambiguous responses. Representative centroid and boundary exemplars from the resulting clusters were then used to construct few-shot prompts for the Llama 3.3 70B model, which generated feedback for the remaining 93 responses. Six independent mathematics instructors evaluated the AI-generated feedback using a structured 0–5 usability rubric. Across all instructor evaluations, 538 of the 558 instructor ratings (96.42%) were 3–5, representing feedback ranging from fair, requiring moderate edits, to excellent, ready to send. More specifically, 482 of the 558 instructor ratings (86.38%) were scores of 4–5, indicating feedback requiring no edits or only minor revisions. The remaining 20 ratings (3.58%) were scores of 0–2, while 13 unique feedback messages received at least one low rating. Across all instructor evaluations, 20 of 558 ratings (3.58%) were assigned scores of 0–2, while 13 unique feedback messages received at least one low rating. Qualitative analysis of low-scoring cases revealed recurring failure modes, including hallucinated completeness in concise solutions, failed arithmetic verification, and false logic flagging for atypical reasoning patterns. These findings suggest that clustering-based exemplar selection may reduce manual prompt-engineering effort while supporting usable LLM-generated formative feedback in a controlled mathematics setting. However, the present study does not compare clustering against alternative exemplar-selection strategies, and therefore conclusions should be interpreted as evidence of feasibility rather than comparative superiority. Full article

Scientific production is expanding so quickly that research teams are struggling to track advances beyond their immediate specialization, especially in interdisciplinary areas where relevant work is scattered across venues and vocabularies. To reduce this overload at the group level, we propose an end-to-end pipeline that transforms structured bibliographic metadata into actionable topic recommendations for research teams. Starting from Scopus records, the method normalizes scholarly text, builds semantic author profiles using Sentence–BERT representations coupled with interpretable keyword descriptors, and forms candidate groups from co-authorship signals and profile similarity. For each group, the approach applies embedding-based topic modeling to generate candidate themes and ranks them using a relevance–novelty trade-off, enabling teams to surface directions that remain aligned with their collective agenda while still encouraging exploration beyond dominant or highly popular topics. Empirical evidence on a Scopus-derived corpus shows that embedding-aware descriptors support cleaner, semantically faithful representations than frequency-based baselines, strengthening downstream topic discovery and producing compact topic lists that are easier for teams to inspect, discuss, and adopt in collaborative planning. Full article

Urban infrastructure systems increasingly depend on textual reports generated by citizens, inspection teams, maintenance units, emergency platforms, and smart city services. Accurate identification of critical risks in these reports is essential for enhancing urban resilience and enabling timely decision-making. Nevertheless, urban infrastructure risk classification is challenging due to the brevity, noise, domain specificity, and context dependence of these reports. This study introduces RAFE-XAI, a retrieval-augmented feature engineering and explainable natural language processing framework for urban infrastructure risk classification. The term retrieval-augmented is used here in a classification-oriented sense: retrieved reports are used to construct additional features and evidence, not to generate output text as in Retrieval-Augmented Generation systems. The proposed framework incorporates semantic sentence embeddings, retrieval-based evidence, neighborhood-derived label distributions, domain-specific risk indicators, infrastructure asset cues, location indicators, and evidence-based explainability. The framework does not construct an explicit graph, adjacency matrix, graph neural network, or message-passing mechanism. Instead, retrieval is used to derive neighbor label-distribution features, which are combined with semantic embeddings and interpretable keyword, asset, and location indicators. To assess the effectiveness of this approach, UIR-Text, a semi-synthetic urban infrastructure risk dataset with scenario-level group splitting to mitigate data leakage, was constructed. Experimental results on UIR-Text show that fine-tuned DistilBERT achieves the strongest predictive performance, with Macro-F1 scores of 0.8278 for category classification, 0.9120 for binary critical-risk detection, and 0.3379 for four-level severity classification. Among the explainable feature-engineering models, RAFE-XAI with Random Forest achieves the strongest category classification performance, with Accuracy 0.8400, Macro-F1 0.8043, Weighted-F1 0.8444, and MCC 0.8062. These results suggest that fine-tuned transformers provide the highest predictive performance on this benchmark, while RAFE-XAI offers a transparent retrieval-augmented alternative that exposes retrieved evidence, neighbor label distributions, and domain cues. Four-level severity classification remains challenging, even with fine-tuned DistilBERT, indicating the need for richer impact-aware variables. Full article

Background: Brain tumor magnetic resonance imaging (MRI) reporting and tumor segmentation for treatment planning are time-consuming and variable. This retrospective fixed-sequence paired workflow study evaluates whether AI assistance is associated with changes in efficiency, consistency, and reproducibility. Methods: Thirty MRI cases (10 vestibular schwannomas, 10 meningiomas, 10 brain metastases) were assessed. Two neuroradiologists completed diagnostic reporting with and without AI assistance, and two physicians completed tumor delineation with and without AI-generated preliminary contours after a 3-week washout. Results: Reporting time decreased from 42.94 to 27.90 min for Reader A and from 101.04 to 80.47 min for Reader B, corresponding to median paired case-level reductions of 40.39% and 11.51%, respectively; only Reader A reached statistical significance. Sensitivity remained 97.73% and 100.00%, while precision was numerically higher after AI assistance (89.58% to 97.73% and 83.02% to 91.67%). Report-similarity metrics increased across ROUGE-L, BERTScore F1, and Sentence-BERT cosine similarity (all p < 0.001). Contouring time decreased from 54.63 to 4.93 min for Reader 1 and from 184.44 to 44.19 min for Reader 2, with median paired reductions of 100.00% and 87.11%. Dice coefficients were numerically higher after AI assistance (0.81 to 0.87 and 0.83 to 0.87). Conclusions: AI assistance was associated with shorter task-completion times, higher report-similarity metrics, and numerically higher contour-overlap measures. Prospective validation should determine whether these workflow efficiency gains translate into broader clinical benefit. Full article

Structural equation modeling (SEM) studies commonly derive constructs and paths from the manually reviewed literature. Expert judgment remains essential, but incomplete coverage and undocumented selection decisions can make this stage difficult to evaluate. We therefore develop a theory-guided, AI-assisted procedure for generating SEM candidate models from systematic literature evidence. PICOC defines the scope, and queries identify the primary records. Backward citation searching adds foundational studies, whereas forward searching adds recent applications. Sentence-BERT and BERTopic are then used to examine semantic structure. Topic terms, representative documents, concept evidence, and theoretical criteria inform the mapping from topics to candidate constructs. The retained constructs are assigned possible SEM roles and assembled into candidate paths. The result is a documented front-end method for candidate model development, not an empirically validated SEM model. Full article

Plain Language seeks to enhance the clarity and comprehensibility of legal and administrative communication; while Natural Language Processing (NLP) offers promising tools for assessing text complexity, most Plain Language classification studies focus exclusively on English, leaving low-resource languages underexplored. This study presents the first systematic evaluation of transformer-based models for sentence-level Plain Language classification in Hungarian tax administrative texts. We benchmarked zero-shot prompting with GPT-4o against fine-tuned open-weight and proprietary models, including huBERT, XLM-RoBERTa, GPT-4o-mini, and Gemini 1.0 Pro, and contextualized these results against previously established lightweight machine learning baselines based on term frequency-inverse document frequency with a support vector machine (TF-IDF + SVM) and fastText. To address data scarcity, we applied translation-based data augmentation using parallel Hungarian–English corpora. The best-performing model achieved a macro-average F1-score of 0.79. Mid-sized models also delivered competitive results, combining accuracy with feasible inference speed and deployment flexibility. Beyond classification performance, we conducted local and aggregated interpretability analysis based on Shapley-values to identify linguistic patterns influencing model decisions. This revealed alignment with known Plain Language features, such as nominalizations and syntactic complexity, as well as biases introduced by frequent domain-specific terms. Our findings demonstrate that Plain Language classifiers can be effectively adapted to low-resource legal–administrative domains. The results support the development of real-time feedback tools that promote linguistic accessibility and contribute to the broader goal of Access to Justice. Full article

Extracting composition expressions from materials science patent documents is essential for patent document searches. Composition expressions describing a single unit of elements and quantities (e.g., “Al: 0.02% or more and 0.08% or less”) tend to appear clustered together. In such cases, researchers in the field of materials science who conduct patent searches have found that boundary-indicating phrases are effective for searching. However, there are no concrete examples that have implemented this approach, and the validity of this approach has not been evaluated to date. In this paper, we propose a Separator Tag (SEP-tag) framework as an explicit boundary for composition expressions in named entity recognition labels. This allows the named entity recognition model to simultaneously perform entity recognition and pattern boundary learning within a single end-to-end process. Furthermore, we propose a four-axis evaluation framework that extends the conventional single-entity F1 score to evaluate named entity recognition models using SEP-tags. (1) Entity F1 score excluding structural tags, (2) Exact match rate for correct spans, (3) Predicted span pattern extraction F1 score, (4) Pattern extraction F1 score. We conducted evaluations using RoBERTa-base and BERT-base-Japanese on materials science patent datasets in English (10,166 sentences) and Japanese (975 sentences). Experimental results show that training the model with SEP-tags improved the span exact match rate on the English dataset by approximately 59.72 percentage points (from 15.95% to 75.67%), and reduced false positives in pattern extraction to 1/117 (F1 score: 0.0784 → 0.8503). In the Japanese dataset, false positives were reduced to 1/123 (F1 score: 0.0361 → 0.4877). For both languages, the entity F1 score was equivalent to that of the model without SEP-tags (English: $|\Delta \mathrm{F}1|<0.002$, Japanese: $|\Delta \mathrm{F}1|<0.001$), with no significant difference found for any of the 13 labels. These results demonstrate that explicit structural boundary tokens are highly effective for extracting composition expression patterns in domain-specific named entity recognition. Full article

Background: Online communities such as Reddit can reveal critical information needs (CINs) among people who discuss opioid use for chronic or acute pain under conditions of stigma and uncertainty. Methods: We used a mixed-method design to analyze 3428 unique Reddit posts from 1 January to 31 December 2022. Sentence-BERT embeddings and K-means clustering identified eight dominant themes, followed by qualitative analysis of a random subsample of 400 posts. Results: The most prevalent theme was seeking alternatives to opioids for pain relief (28.00%), followed by medications for chronic pain (19.84%) and kratom use experiences (14.64%). Qualitative findings showed recurrent concerns about addiction, stigma in clinical encounters, withdrawal, treatment access, and peer-to-peer discussion of kratom, tianeptine, cannabis, and other substitute substances. Conclusions: The study suggests an unmet need for accessible, nonstigmatizing, and clinically grounded information on pain management, tapering, and lower-risk alternatives to opioid use. These results have implications for addiction prevention, harm reduction, and patient-facing communication strategies for clinical practice and community health. Full article

Aligning academic transcripts with occupational competency requirements remains challenging because course labels and job-skill terms are semantically ambiguous, role-specific, and difficult to explain. This paper proposes the Ontology Framework for Multi-level Competency Mapping (O4CM), a semi-automated framework integrating a Large Language Model (LLM) ensemble, Human-in-the-Loop (HITL) verification, Sentence-BERT (SBERT) semantic representation, the Path Consistency Index (PCI), and Total Accumulated Competency Score/Normalised Total Accumulated Competency Score (TACS/NTACS) ranking. O4CM was evaluated on a historical job-posting corpus and anonymised transcripts from five university programmes through ablation, sensitivity analysis, baseline comparison, and expert-labelled validation. The LLM ensemble reached high consensus for 21 of 22 Occupational Information Network (O*NET) knowledge-domain mappings (95.45%), each of which was subsequently expert-verified. In a computing-only expert-aligned analysis, the full framework most closely matched expert rankings across three data-domain roles. Within this dataset, ontology-path evidence can support more transparent competency ranking for educational advising and exploratory recruitment screening. Full article

Medical text classification is an important task in biomedical natural language processing, but prediction errors remain problematic in high-stakes settings where reliability matters in addition to accuracy. To address this challenge, this paper proposes a learned selective deferral framework for biomedical sentence classification that allows uncertain predictions to be deferred under constrained review budgets. The framework combines a transformer-based classifier with uncertainty estimation, temperature scaling, and a learned deferral policy that predicts the likelihood of model error from multiple signals, including confidence, entropy, calibration-aware features, and Monte Carlo Dropout descriptors. Deferral decisions are applied under fixed budgets to improve the use of limited review capacity. Experiments on the PubMed 200k RCT dataset show that budget-constrained deferral reduces system-level risk. Using PubMedBERT as the primary backbone, deferring 20% of the highest-risk cases reduces system risk from 0.1108 to 0.0360. Compared with a calibrated confidence-threshold baseline, the learned policy provides modest but generally favorable improvements, with statistical significance observed at the 20% budget. Additional experiments across PubMedBERT, BioBERT, and SciBERT suggest that the framework transfers across biomedical transformer backbones, while calibration improves the reliability of confidence estimates and learned policies outperform random deferral. Full article

Disambiguation resolution in speech synthesis is one of the main challenges in text-to-speech conversion. Machine learning methods and artificial neural networks have been successfully applied to this problem in synthesis systems for English, Spanish, and other common languages. For low-resource languages, the available data are insufficient to train artificial neural networks, so heuristic methods for context analysis and selection of the correct homonym for polysemantic words should be used. The purpose of this study is to develop a word sense disambiguation (WSD) method for the low-resource Chechen language and to introduce it into a speech synthesis system. The study presents the developed method and three algorithms: AWEN (based on Euclidean distance), AWA (weighted average), and AWN (weighted normalized distance) for word sense disambiguation. A corpus of Chechen texts, CheWSData, was compiled, containing 15,035 manually selected sentences derived from 5 million annotated words and reflecting the natural frequency of polysemy across grammatical categories. Experimental results show that the proposed AWN method achieves the best performance, with an F1-score of 0.78 and an accuracy of 0.80, outperforming AWA (F1: 0.74) and AWEN (F1: 0.40). For specific parts of speech, AWN reaches F1-scores of 0.82 for nouns, 0.83 for verbs, and 0.85 for adverbs. Comparative analysis with existing WSD methods for low-resource languages (Kashmiri, Hausa, Assamese, Urdu, and Vietnamese) demonstrates that AWN is competitive, ranking second after ViConBERT (F1: 0.87) and ahead of XLM-R for Hausa (F1: 0.79). The developed software module for homonym recognition was integrated into the Chechen speech synthesis system, contributing to more natural synthesized speech. Full article

Manual case review in legal and courtroom workflows faces efficiency bottlenecks. While LLMs offer potential for vertical domains, they often struggle with domain-specific accuracy and hallucinations. This paper introduces ILL, a lightweight model for legal and courtroom assistance trained via QLoRA. By employing 4-bit quantization on an RTX 4060 GPU, ILL achieves precise knowledge transfer with low computational costs. The model attained a BertScore F1 of 0.8037, and a perplexity of 1.89, while largely preserving TruthfulQA performance after fine-tuning and demonstrating competitive results on MMLU tasks. Experiments demonstrate that this method performs excellently on small-scale datasets and shows approximate convergence at scales of about 1200 sentences. This work validates the feasibility and efficiency of constructing high-quality vertical auxiliary models using limited computational resources. Full article