Search Results (354)

Open workforce datasets can help to extend research on artificial intelligence (AI) only if they are sufficiently provenance-traceable, have good measurement quality, and have a relational structure suitable for behavioral inference. This study examines a benchmark dataset comprising 12,000 linked records across 15 industry categories and 47,206 AI tool-use records. Sampling, recruitment, questionnaire wording, respondent authentication, ethics procedures, and whether the records are real or synthetic are not documented, so the dataset is treated as a tabular source rather than verified workforce evidence. Analyses are limited to indicators that have been observed directly, including job satisfaction, work–life balance, career outlook, trust in AI, weekly learning hours, AI-use intensity, and employer-provided AI training. Assessment of behavioral interpretation of the data was conducted using Spearman correlations, HC3-robust regressions, false discovery rate adjustment, secondary industry interaction checks, and a relational-realism diagnostic. Concerns about AI replacing jobs were negligible and non-significant on a five-point scale, with an average of 2.745. Training also showed no robust association when evaluated against outcomes that did not contain training. The median absolute Spearman correlation across conceptually related observed variables was 0.0064, with the 95th percentile at 0.0180, which is very low. These estimates reflect the characteristics of the supplied documents rather than employee actions, and do not assess conservation of resources processes or tourism worker outcomes. Results from the analysis demonstrate the importance of checking open workforce data for construct validity, relational validity, sector fit and provenance before testing behavioral theories. Research for regenerative tourism requires reliable sector-specific samples, reliable multi-item measures, longitudinal design, and direct measures of social and destination outcomes. Full article

Online social networks generate large volumes of textual data that reflect users’ opinions, affective expressions, and broader patterns of engagement and social behavior. However, natural language processing approaches frequently examine sentiment, trust-related signals, and behavioral indicators independently, limiting their ability to represent the multidimensional nature of online interaction. This study conducts a systematic comparative evaluation of lexicon-enhanced fine-grained sentiment classification using linguistic, message-level statistical, and lexicon-derived affective information across a common experimental framework. The empirical analysis combines TF–IDF features, word-count information, and sentiment indicators derived from TextBlob, SentiStrength, and VADER, while the broader multi-level organization is used to relate the resulting affective evidence to online social-behavior analysis. Fifteen classical machine learning algorithms and seven deep learning architectures are evaluated on a real-world Twitter dataset containing 41,157 COVID-19-related tweets labeled across five sentiment-intensity classes. The experimental evaluation considers four feature configurations and seven performance metrics, complemented by Friedman and post hoc Wilcoxon signed-rank tests. The results show that TextBlob provides modest improvements, SentiStrength produces broader and more consistent gains, and VADER yields the strongest overall performance. AdaBoost combined with VADER achieves the best results, with 93.16% accuracy, 93.20% macro F1, 93.27% balanced accuracy, and an MCC of 0.913, while the Dense Neural Network is the strongest deep learning model. These results demonstrate that lexicon-derived affective features can substantially strengthen fine-grained sentiment classification, although their effectiveness depends strongly on the learning algorithm used to exploit them. The empirical contribution of this study is confined to fine-grained sentiment classification, while trust-related and attachment-related dimensions are retained as higher-order interpretive constructs rather than directly predicted or empirically validated outcomes. Full article

Adaptive and explainable systems are two critical concepts in personalized learning that enhance the customized experience based on the adjusted content according to the individual learning of a learner. In this paper, we present a modular and interpretable architecture for the development of a System for Adaptive and Explainable Feedback, which is a potential solution for the above-mentioned issues in personalized learning environments. The adaptive feedback provided by the System for Adaptive and Explainable Feedback is in real time. Still, even more importantly, it is justified clearly and transparently so that both the learner and instructor understand the why behind the recommendations for action given by the system. It consists of modular technologies for data collection (word and phrase occurrence, time tracking), analysis (latency, process mining), feedback generation (adaptive after-action review), and result presentation, making a modular, flexible, and scalable system to suit many educational scenarios. To evaluate the system’s functional performance, the SAEF pipeline was applied to a dataset derived from the ASSISTments platform, a well-established educational dataset widely used in learning analytics research. On a cohort of 500 student profiles, the SAEF achieved an overall recommendation accuracy of 83.6%, a weighted F1-score of 82.9%, and a mean system response time of 42.1 ms, demonstrating both the internal computational consistency and efficiency of the adaptive pipeline. These results indicate strong agreement with score-derived difficulty categories and support the internal computational consistency of the recommendation pipeline as a proof-of-concept system, though they do not constitute independent evidence of instructional appropriateness. The SAEF is designed to support learner engagement, personalize learning pathways, and foster transparency in AI-driven educational environments; a full empirical evaluation involving real-world deployment is identified as the primary direction for future work. The perceived understandability, trustworthiness, and pedagogical usefulness of the SAEF’s explanations by learners and instructors represent a complementary dimension yet to be empirically explored. The SAEF’s architecture is designed with the explicit objective of supporting learner engagement and fostering transparency; however, these pedagogical benefits are architectural design goals rather than empirically demonstrated outcomes in the present study, which focuses exclusively on computational validation. Full article

Background: WordPress plugins account for the large majority of disclosed CMS vulnerabilities, and learning-based detectors report high accuracy on synthetic corpora and random splits. Methods: We build a benchmark from 1757 real plugin CVEs (6666 indexed identifiers, 1281 plugins), yielding 30,860 labeled PHP functions in two negative-sampling variants. An audit of the released artifacts found exact normalized-code duplicates crossing test folds (166 groups/1086 records in Variant A; 151/676 in Variant B) and identical code carrying contradictory labels (730 and 456 groups). After removing conflicting groups, merging exact duplicates and collapsing token- and AST-level clone families, the corpora contain 15,294 (Variant A) and 24,917 (Variant B) records with zero shared hashes, clone families or plugins across folds. Ten detectors are compared under plugin-grouped five-fold cross-validation on complete held-out folds: a static taint analyzer, TF-IDF Random Forest and SVM, Bi-LSTM, unidirectional LSTM, a taint-augmented neural model, a nested out-of-fold stacking ensemble, and three baselines (majority class, a source/sink presence rule, and a single-feature function-length classifier). Results: The best macro-F1 is 0.563 (Variant A, binary, SVM) and 0.601 (Variant B, binary, nested stacking). Neural models do not lead in any of the six configurations, and the single-feature length baseline matches or exceeds the TF-IDF Random Forest in four of six. The source/sink rule attains an MCC close to zero, and the taint analyzer misses about 98% of vulnerabilities. Manual review of 400 functions finds 48.8% of cleaned positive labels to be genuine vulnerabilities. Conclusions: Under duplicate-controlled evaluation, the differences between detector families are small compared with the variation attributable to label quality, which appears to be the more binding constraint. All data, folds, predictions and fingerprinted result files are released. Full article

Machine learning has become a widely adopted approach for URL-based phishing detection, with many studies reporting F1 scores exceeding 0.95 on benchmark datasets. However, recent adversarial machine learning research has questioned the robustness of these models, suggesting that small input perturbations can severely degrade detection performance. In this study, we argue that a substantial part of this reported vulnerability stems from the way adversarial attacks are evaluated. Specifically, many existing studies assess attacks in the feature space, where feature values are modified directly without ensuring that the resulting samples correspond to valid, functional URLs. To investigate this issue, we conduct a two-stage empirical study using both a benchmark feature dataset and a dataset of real phishing URLs. Crucially, to avoid confounding the attack space with dataset differences, we additionally evaluate both feature-space and problem-space attacks on the same real-URL dataset, using an identical model and manipulable-feature budget. Our experiments reveal a striking contrast between these evaluation settings. While feature-space attacks reduce the detection rate of a Random Forest classifier on the benchmark dataset from 0.96 to 0.36, analogous manipulations performed on real URLs have almost no effect on detection performance, as the most informative signals originate from host-related attributes that are difficult for attackers to manipulate. Building on this observation, we propose a set of robust features that capture stable domain characteristics, including lexical word validity, homoglyph disguises, brand impersonation, subdomain depth, character entropy, and transport-related signals. Incorporating these features substantially improves robustness under adversarial conditions, maintaining phishing detection rates between 0.24 and 0.76 where the lexical-only baseline deteriorates to zero under a non-adaptive attacker, while also increasing the clean-data F1 score from 0.985 to 0.994. We further evaluate an adaptive attacker that explicitly targets the proposed features; although the proposed representation raises the attacker’s cost and helps under moderate attacks, host-derived features remain the only strictly attack-invariant component, so we position the proposed features as a complement to host-based signals rather than a standalone defense. Additional analyses, including model comparison, hyperparameter sensitivity analysis, feature ablation, SHAP-based interpretation, multi-seed confidence intervals, a domain-disjoint evaluation, and host-only evaluation, consistently support the proposed approach. The findings demonstrate that problem-space evaluation provides a more realistic assessment of adversarial robustness than conventional feature-space testing and show that robust feature engineering offers a practical strategy for developing phishing detection systems that remain effective under realistic adversarial conditions. Full article

Degraded tax forms and other legal–administrative documents require restoration methods that impose strict constraints on content fidelity. This paper presents a structure-aware restoration framework for degraded tax documents that combines geometric normalization, canonical template guidance, and supervised image restoration within a common alignment space. Documents are first mapped to a canonical layout through homography estimation based on Scale-Invariant Feature Transform (SIFT) and Random Sample Consensus (RANSAC), using the canonical visual template as the reference. Restoration is then performed using template priors and masked constraints designed to preserve the fixed document structure while recovering variable content. Under a fixed-budget comparative protocol, U-Net with template priors achieved the highest visual and structural quality, whereas the proposed hybrid model obtained the best functional OCR performance on synthetic data, reaching a Character Error Rate (CER) of 0.1091 and a Word Error Rate (WER) of 0.3495. As a complementary evaluation on 37 real-world documents with human-generated textual ground truth, restoration increased full-page word coverage across all three OCR engines evaluated, yielding absolute improvements ranging from 4.15 to 18.05 percentage points. Full article

Architectural atmosphere for spirituality holds potential to support the physical and mental health of post-COVID urban populations, and people now spend more of their lives within buildings even as opportunities for spiritual engagement recede. Yet evidence-based knowledge for designing such environments remains limited. This study aimed to identify the architectural atmospheres for spiritual constructs through a systematic literature review and practice-based interviews conducted following the COVID-19 pandemic. Data from Scopus-indexed publications and architects’ relevant expertise were analyzed using thematic content analysis, while word frequency and bivariate association analyses were employed to identify key architectural characteristics and their associations with spiritual constructs. The findings indicate that a holistic atmospheric approach should integrate spatial, perceptual, and natural features. Tangible atmospheric features such as space, natural materials, and object elements emerged as most prominent, alongside intangible features including illumination, acoustics, and temperature conditions. At the use-effect level, spatial experience and perception were most frequently reported. Positive spiritual constructs were linked to symbolic objects, sensory experiences, and lighting conditions, whereas negative spiritual constructs were primarily associated with temperature conditions. Future studies should extend inquiry across diverse linguistic, cultural, and temporal contexts while empirically validating the proposed framework in real-world environments. Full article

As the construction of new-type power systems continues, the extensive integration of renewable energy and power-electronic devices has significantly increased the risk of cascading failures in power grids, making accurate prediction of cascading-failure evolution paths crucial for grid security. However, real-world fault samples are extremely scarce, and conventional physics-based simulations are too computationally intensive for real-time online early warning. To address these challenges, this paper proposes a cascading failure evolution path prediction method based on massive event chain mining and prior knowledge constraints. First, we construct a refined cascading-failure simulation model that incorporates the action logic of the three defense lines to generate a large standardized event-chain dataset, thereby addressing the data scarcity faced by data-driven models. Second, a sequence-prediction engine that combines word embeddings, LSTM-based temporal modeling, and prior-knowledge constraints is developed; the physical action logic of the power system is explicitly incorporated into the loss function to improve the physical plausibility of the predictions. Finally, an evaluation framework from event prediction to third-defense-line early warning is constructed to assess system-level security risks. Based on simulations of the IEEE 39-bus AC/DC hybrid system, the proposed method is shown to achieve single-step Top-1 accuracies of 97.0% and 96.8% for the system-level critical events of frequency limit violation and system instability, respectively, with corresponding Top-3 accuracies of 99.2% and 99.4%. The overall Top-1 and Top-3 accuracies are 89.8% and 98.2%, respectively; the mean single-inference time is approximately 9.6 ms, corresponding to a speedup of more than 4700 times over conventional time-domain simulation, and the weighted mean early-warning lead time is 1.78 s. Full article

Purpose: This study conducts a systematic literature review on the intersection of real estate exposure and geotechnical hazards, focusing specifically on seismic and subsurface risks. The objective is to synthesize key thematic trends, methodologies, and governance frameworks that inform risk-informed planning in seismically vulnerable urban areas. Design/methodology/approach: A Boolean search query was implemented on Lens.org, identifying 55 peer-reviewed articles published between January 2020 and May 2025. Inclusion criteria required explicit focus on property exposure to seismic or ground instability risks. Thematic analysis was conducted based on title and abstract data, supported by a Python (version 3.11)-generated word cloud to inductively identify five core clusters: (1) seismic assessment and earthquake risk, (2) building vulnerability and structural performance, (3) subsurface hazards and ground instability, (4) urban areas, heritage, and social vulnerability, and (5) risk mitigation, planning, and resilience frameworks. Findings: The review reveals a shift from hazard-centric, engineering-based models toward integrated, multi-scalar frameworks that embed risk within socio-economic, spatial, and institutional contexts. While consensus exists on the importance of probabilistic modeling, retrofitting, and GIS-based tools, divergences persist around behavioral valuation, policy uptake, and equity in implementation. Heritage cities and informal settlements emerge as under-addressed but critically vulnerable domains. Originality/value: This study systematically maps interdisciplinary research on real estate exposure to seismic and subsurface risks post-2020. By bridging engineering, planning, behavioral economics, and disaster governance, the review provides a unique synthesis relevant for academics, urban planners, and policymakers seeking to design equitable and resilient urban futures. The five-cluster thematic taxonomy introduced in this review represents an original synthesis that bridges engineering vulnerability assessment, behavioral economics, heritage preservation, and resilience governance. Unlike previous reviews that have typically focused on single disciplinary perspectives, this taxonomy integrates multi-scalar approaches spanning asset-level diagnostics to national exposure modeling, providing a comprehensive framework for understanding real estate exposure to seismic and subsurface risks. Full article

A sign language interpretation system aims to translate sign gestures into spoken or written language in real time, enabling signers and non-signers to communicate in their familiar linguistic forms. However, vision-based approaches suffer from hand occlusion, lighting variability, and complex backgrounds, while Deep Neural Network (DNN)-based methods incur heavy computational costs that hinder real-time use on resource-constrained platforms. In this paper, we propose sign language gloves and a lightweight motion retrieval method for real-time sign language interpretation that runs on mobile devices and embedded systems. The sign language gloves integrate flex sensors, an inertial measurement unit (IMU), and pressure sensors to accurately capture gesture features, including finger bending angles, hand orientation, movement trajectories, and fingertip contacts with body parts, enabling recognition of touch-based gestures. For the motion retrieval method, we build a comprehensive gesture dataset with the gloves and perform feature analysis on each sign language gesture to avoid redundant information in the dataset. During interpretation, our system employs a feature-labeling mechanism to ensure gesture distinguishability and a gesture retrieval algorithm to evaluate movement continuity and similarity. This allows the system to identify corresponding feature labels and consolidate them into complete sign language vocabulary entries. The proposed motion retrieval method is characterized by low computational complexity and a well-defined data structure. This makes it suitable for integration into embedded systems, offering real-time performance and high portability for practical deployment. In our experiments, the proposed system achieved an average recognition accuracy of $92\%$ on a gesture dataset covering 300 sign language words. Full article

Semantic radicals are meaning-bearing sublexical components of Chinese characters that have been shown to facilitate incidental vocabulary learning in L1 Chinese readers. However, it remains unclear whether this sublexical mechanism extends to learners of Chinese as a foreign language (CFL), and how semantic radicals shape the online processing and learning outcomes of novel words in both populations. The present study investigated the role of semantic radical consistency in incidental vocabulary learning among L1 readers and CFL learners. Eye movements were recorded while participants read sentences containing novel two-character pseudowords whose semantic radicals were manipulated to be either consistent or inconsistent with the word’s semantic category. Both groups successfully acquired orthographic and semantic representations of the novel words, with L1 readers outperforming CFL learners. Eye-tracking data revealed a significant group-by-consistency interaction during initial exposures. L1 readers exhibited dynamic sensitivity to radical consistency, allocating more fixation time to inconsistent items at first exposure and rapidly adjusting by the second, whereas CFL learners showed no reliable online consistency effects. Moreover, reading time proved a task-specific predictor of learning outcomes. Longer initial fixation benefited orthographic learning, while extended cumulative reading time was negatively associated with semantic accuracy. Taken together, these findings suggest that sublexical semantic cue deployment during real-time reading is qualitatively distinct between L1 readers and CFL learners. Full article

Pedestrian Spatial Experience PSE in urban spaces is a multi-faceted topic that requires the thematization of assessment methods due to their fragmentation across studies. Accordingly, this systematic review followed an inductive approach to define a framework of PSE assessment themes reflecting their evaluation in public and university campus spaces. This systematic review included open-access, accessible, peer-reviewed sources based on assessment-focused English research that followed defined frameworks on the effects of urban environments on adult PSE. Studies were excluded if they focused on non-pedestrians or vulnerable user groups, examined non-pedestrian-scale contexts, explored pedestrian experience in virtual environments, assessed interior spaces, lacked a structured attribute-based assessment framework, were review articles, did not specify how urban environments shape pedestrian experience, investigated non-urban or rural areas, or examined urban settings without clearly defined street or square infrastructure. The review relied on querying PSE-related bibliography from the Scopus and Web of Science databases on 12 October 2025; results were processed through a screening procedure according to the inclusion and exclusion criteria. The final sets of sources reviewed included 83 and 24 sources related to PSE assessment in public and university campus spaces, respectively. Risk of Bias (RoB) tools included the Joanna Briggs Institute (JBI) tool for cross-sectional studies, tailored for urban spatial studies, and the Prediction model Risk Of Bias Assessment Tool (PROBAST+AI), tailored for ABM studies. Using a data extraction sheet and codebook to identify the prominent codes in the included sources, in addition to reviewing frequent words and the methods of the included sources, clarified the main conceptual framework of PSE assessment themes. The thematic categorization of PSE studies was followed by analyses of the frequencies of the themes, the prevalence of themes across countries and cities, and the theoretical explorations within the themes over the years in both reviewed contexts. Subsequently, synthesizing both sets clarified the interrelations between themes, methods, and tools as an attempt to address gaps in PSE assessment methods. The main results of this review are the 11 themes of PSE assessment that were identified from the reviewed sources. Data analyses and syntheses indicated a high prevalence of quantitative methods relying on visual aspects, signifying the dominance of the Cognitive and Navigational Experience theme due to its frequent assessment by numerous and diverse sets of methods in both reviewed sets. Nevertheless, the Temporal Experience theme emerged as the least considered. The key limitations of this systematic review include its reliance on accessible articles from bibliographic databases, as well as its focus on adult populations as the common users of public and university campus spaces. This review decodes PSE in terms of its assessment themes through the methods followed and the applied tools within real environments. As an application of the introduced conceptual framework, this systematic review clarifies the comparison of the themes examined between public and university campus spaces. The findings of this systematic review provide a foundation for a comprehensive understanding of PSE, thereby informing the design of more user-centered environments. Full article

(1) Background: Clinical trial data extraction from registries such as ClinicalTrials.gov remains labor-intensive and error-prone, often missing critical details hidden in unstructured protocol descriptions. Large Language Models (LLMs) offer potential to automate this process, yet systematic multi-model comparisons on real clinical trial data remain scarce. (2) Methods: Four LLMs (OpenAI o4-mini-high, Anthropic Claude-Sonnet-4, Google Gemini 2.5-Pro, and Meta Llama-4-Maverick) extracted brain stimulation parameters from 67 transcranial direct current stimulation (tDCS) trials in Parkinson’s disease via a structured JSON schema. Pairwise inter-model agreement was quantified with Cohen’s Kappa and percentage agreement across binary, categorical, and multi-component task tiers. (3) Results: Under exact-string matching, agreement was near-perfect for binary classifications (non-invasive classification: 100%; brain stimulation presence: 99.3%, $\kappa$ = 0.50) and substantial for categorical extractions (primary stimulation type: 96.4%, $\kappa$ = 0.70), but fell to 48.6% ($\kappa$ = 0.43) for complex anatomical targets. Numeric parameters revealed model-specific strengths: o4-mini-high and Claude-Sonnet-4 achieved perfect duration agreement (r = 1.000, n = 19) while Llama-4-Maverick diverged substantially (r < 0.12). Validation against an expert gold standard (100% inter-annotator agreement on a 20-trial overlap) confirmed high extraction accuracy across all features (mean 93.7–98.9%). Crucially, the low agreement on anatomical targets proved to be an artifact of exact-string scoring: under the same semantic matching used to measure accuracy, inter-model agreement rose to 97.0%, coinciding with the 95.5% expert accuracy. Inter-model agreement therefore tracks accuracy once both are measured on a common basis. (4) Conclusions: Exact-string inter-model agreement decreases with task complexity, but this decline largely reflects interchangeable free-text wording rather than reduced accuracy. Evaluated semantically, agreement and expert accuracy are both high and closely aligned. A residual risk is not low accuracy but the rare error shared across all models, which agreement cannot detect, and which overall accuracy can itself mask when one class dominates. These findings inform hybrid human–AI systematic review pipelines in which targeted expert oversight focuses on shared-error and minority-class detection. Full article

Urban renewal increasingly adopts Nature-Based Solutions (NBSs) to address environmental challenges and enhance social well-being. However, it remains unclear whether and to what extent NBSs contribute to cultural ecosystem services (CESs), which reflect people’s perceptions, values, and experiences of urban nature. This study develops an integrated framework combining text and image mining of social media data to evaluate the CES outcomes of NBS in regenerated urban districts in Chengdu, China. The comment data were analyzed for CES using Jieba word segmentation and dictionary matching, while images were categorized into NBS types by manual classification. By integrating these multimodal data, the framework effectively clarifies the relationship between NBSs and CESs from the perspective of public perception. Results indicate that recreation and leisure, inspiration, and spiritual values are the most prominent aspects of public perception, with linear green infrastructure and pocket parks being the most frequently identified NBS types. Correspondence analysis further reveals significant associations between specific NBS interventions and CES categories. By integrating textual and visual data, this study offers a practical and real-time approach for capturing public perceptions of CESs and provides actionable insights for the design and management of NBS-driven urban regeneration. Full article

With the rapid growth of Chinese social media data, many language-driven analytical tasks, such as sentiment analysis and malicious account detection, are increasingly formulated as computationally expensive optimization problems, particularly in the context of hyperparameter tuning for deep learning models. Due to the intrinsic characteristics of Chinese text, including implicit word boundaries, strong context dependency, and high linguistic variability, the resulting feature representations are often high-dimensional, sparse, and heterogeneously distributed. From an optimization perspective, these properties induce highly irregular, non-smooth, and multimodal objective landscapes, posing significant challenges to conventional surrogate-assisted data-driven evolutionary algorithms (DDEAs). To address this problem, this paper proposes a Normal Selection-based data-driven evolutionary algorithm (NSEA) for improving surrogate-assisted optimization under complex conditions. Specifically, a Normal distribution-based selection strategy (NSS) is developed to enable probabilistic selection of surrogate models, balancing exploitation of high-performing models and exploration of alternative candidates, thereby alleviating premature convergence in multimodal search spaces. In addition, an exponential weighting ensemble (EWE) method is introduced to aggregate surrogate models based on their relative ranking performance, which enhances the stability and generalization capability of fitness approximation across different regions of the search space. Extensive experiments on benchmark functions demonstrate that the proposed NSEA consistently outperforms several state-of-the-art DDEAs in terms of optimization accuracy and robustness. Furthermore, a real-world application of cheating official account (COA) detection on Chinese social media is conducted, in which the hyperparameter optimization of a heterogeneous graph transformer (HGT) model is formulated as an EOP. The results further prove the effectiveness and practical applicability of the NSEA in complex data-driven scenarios. Overall, this study provides an effective optimization framework for handling EOPs with complex and multimodal characteristics and offers a feasible computational approach for tasks associated with large-scale Chinese textual data. Full article