Search Results (24,568)

Metadata extraction from natural history collection labels is a pivotal task for the online publication of digitized specimens. However, given the scale of these collections—which are estimated to host more than 2 billion specimens worldwide, including ca. 400 million herbarium specimens—manual metadata extraction is an extremely time-consuming task. Thus, automated data extraction from digital images of specimens and their labels therefore is a promising application of state-of-the-art computer vision techniques. Extracting information from herbarium specimen labels normally involves three main steps: text segmentation, multilingual and handwriting recognition, and data parsing. The primary bottleneck in this workflow lies in the limitations of Optical Character Recognition (OCR) systems. This study explores how the general knowledge embedded in multimodal Transformer models can be transferred to the specific task of herbarium specimen label digitization. The final goal is to develop an easy-to-use, end-to-end solution to mitigate the limitations of classic OCR approaches while offering greater flexibility to adapt to different label formats. Donut-base, a pre-trained visual document understanding (VDU) transformer, was the base model selected for fine-tuning. A dataset from the University of Pisa served as a test bed. The initial attempt achieved an accuracy of 85%, measured using the Tree Edit Distance (TED), demonstrating the feasibility of fine-tuning for this task. Cases with low accuracies were also investigated to identify limitations of the approach. In particular, specimens with multiple labels, especially if combining handwritten and typewritten text, proved to be the most challenging. Strategies aimed at addressing these weaknesses are discussed. Full article

In the past decade, a growing number of cyberattacks have been reported, enabling unprecedented levels of personalization, automation, and deception. For instance, recent industry surveys have reported sharp increases in unique social engineering attacks within a single month of 2023, coinciding with the public release of ChatGPT-3.5. This trend highlights how Artificial Intelligence (AI)-powered phishing campaigns have become a significant threat to digital ecosystems. The present study provides an integrative analysis of how generative and deepfake technologies have reshaped the landscape of a Social Engineering (SE) attack, categorizing the main attack strategies and examining their psychological, technological, and ethical implications. In addition, to reviewing enabling technologies, our study conducts a comparative analysis of frameworks and analytical models across technical, empirical, and quantitative perspectives that model AI-driven SE operations and their defensive countermeasures. The convergence of these frameworks reveals three core capabilities—realism, personalization, and automation—that systematically amplify attack efficiency. Building on these insights, the study proposes the Unified Model for AI-Driven Social Engineering (UM-AISE), a conceptual framework that integrates these dimensions across the attack lifecycle and employs a theoretical Markov Decision Process (MDP) analysis. This formalization demonstrates how these capabilities can shift the attacker’s optimal strategy, offering a formal economic perspective distinct from empirical validation. Finally, the study discusses emerging ethical and regulatory challenges associated with AI-mediated deception, highlighting risks related to opacity, accountability, and large-scale manipulation. Taken together, these elements inform evolving approaches for detection, defense, and governance relevant to researchers, policymakers, and practitioners. Full article

Hazard recognition on construction sites is crucial for ensuring worker safety. Traditional methods widely rely on expert assessments, on-site inspections, and checklists, which can be time-consuming and susceptible to human error. The integration of multimodal Large Language Models (LLMs), such as GPT-based systems, offers a promising opportunity to overcome these limitations. Therefore, this study evaluates the effectiveness of GPT-4o in recognizing workplace hazards from image inputs, with a specific focus on construction sites. The results indicate that the model can serve as a valuable decision-support tool for safety professionals by providing scalable and real-time insights. However, the study also highlights key limitations, including the model’s reliance on general visual features rather than domain-specific safety knowledge, and the continued need for human supervision. Additionally, ethical concerns, including bias in AI-generated hazard assessments, data privacy, and the risk of over-reliance on AI, must be carefully managed to ensure these tools contribute responsibly and effectively to proactive risk management strategies. Full article

Background: This integrative review investigates how behavioural and psychological factors shape non-market environmental valuation within the scope of sustainable development. Unlike traditional technical-economic approaches, the novelty of this work lies in reframing socio-cultural drivers of pro-environmental behaviours (PEBs) within macro sustainability paradigms and proposing a socially and ethically grounded framework. The review has three objectives: (i) to incorporate psychological and socio-cultural dimensions into the sustainable development agenda; (ii) to demonstrate how values, norms, and perceptions drive PEBs; and (iii) to call for an ethical consensus across socio-economic and environmental sustainability. Methods: The review follows PRISMA 2020 guidelines and synthesises English-language empirical and conceptual studies (2010–2025) from Scopus and Web of Science, supplemented by Google Scholar. The literature search was conducted in December 2025, and rigorous screening and exclusion criteria were applied to ensure methodological reliability. Results: The review includes 69 interdisciplinary studies and 2 reports. The synthesis yields a framework on ethics that integrates psychological, behavioural, and economic perspectives in non-market environmental valuation and informs the weak vs. strong sustainability debate. Discussion: The findings connect sustainability debates to socio-cultural theories to explain how values, norms, and perceptions shape PEBs and valuation-relevant preferences. The review is limited by its integrative (non-meta-analytic) design, which relies on qualitative synthesis and expert judgement across heterogeneous theoretical and empirical traditions; therefore, a formal risk-of-bias assessment was not conducted. The review protocol was registered on OSF (registration ID W9Y8T). Full article

Background: Periapical pathosis in periapical radiographs must be properly diagnosed for the success of endodontic treatment but is often muddled by 2D imaging limitations and subjective interpretation. Artificial intelligence (AI) offers a solution, but whether the diagnostic granularity of AI versus human clinicians in everyday clinical practice has been adequately explored remains to be addressed. The purpose of this study was to evaluate the diagnostic accuracy of ChatGPT-5 in detecting periapical radiographic abnormalities compared with the three-expert consensus reference standard. Methods: In this diagnostic accuracy retrospective study, 270 periapical radiographs were independently read by a large language model (ChatGPT-5) and a three-board-certified oral radiologist consensus. The AI was given a standardized prompt to label radiographic features, like the presence of periapical radiolucency, border, shape, and integrity of lamina dura. Diagnostic accuracy, agreement (Cohen’s κ), and predictors of correct AI classification were compared with the expert consensus reference standard. Results: ChatGPT-5 demonstrated high sensitivity (87.5%) but low specificity (12.5%), resulting in an overall diagnostic accuracy of 50.0%. This performance profile reflects a tendency toward over-identification of pathology, with the model classifying 87.5% of radiographs as abnormal compared with 50.0% by expert consensus. Agreement was almost perfect for anatomical localization (arch, κ = 0.857) but poor for binary abnormality detection (κ = 0.000). For morphological descriptors, statistically significant disagreement was observed for lesion border characterization (κ = 0.127; p < 0.001), whereas lesion shape demonstrated only descriptive divergence without reaching statistical significance (κ = 0.359). Root resorption assessment also differed significantly between evaluators (p = 0.046). Regression analysis showed that well-defined corticated borders (OR = 60.25, p < 0.001) and first molar-associated lesions (OR = 32.55, p < 0.001) were significant predictors of correct AI classification. Conclusions: This study demonstrates that while ChatGPT-5 Vision can visually interpret periapical radiographs with high sensitivity, limited specificity and inconsistent morphological feature characterization restrict its reliability for independent clinical diagnosis. The AI system tends to over-diagnose systematically and categorizes lesions more structurally and defined compared to dental experts. AI has the potential for being optimized as a sensitive first-screening test, but its findings must be validated by dental professionals to avoid false positives and ensure proper characterization. Full article

A company’s reputation is an important, intangible asset, which is heavily influenced by media reputation. We developed a method to measure a company’s reputation based on sentiments detected in online articles. The sentiment of each sentence was evaluated and categorized into one of three polarities: positive, negative, or neutral. Then, we developed another method to assess a company’s media reputation using all available online articles about the company. The company’s media reputation is presented as a tuple consisting of their media reputation on a scale from 0 to 100, the number of articles related to the company, and the margin of error. Experiments were conducted using articles written in Lithuanian published on major news portals. We used two different tools to assess the sentiments of the articles: Stanford CoreNLP v.4.5.10, combined with Google API, and the pre-trained transformer model XLM-RoBERTa. Google API was used for translation into English, as Stanford CoreNLP does not support the Lithuanian language. The results obtained were compared with those of existing methods, based on the coefficients of media endorsement and media favorableness, showing that the results of the proposed method are less moderate than the coefficient of media favorableness and less extreme than the coefficient of media endorsement. Full article

This article adopts a comparative mythology framework in order to situate creation myths within a broad cultural context. It examines how different societies conceptualize the emergence of the universe through the interconnected notions of thought and word. The study demonstrates that, despite cultural diversity, these narratives articulate shared principles concerning the mental and linguistic foundations of existence while preserving tradition-specific expressions. The analysis is based on qualitative content analysis of primary mythological texts drawn from Hindu, Maori, Maya, Maiana, Dogon, Polynesian, Ancient Egyptian, and Turkish traditions, encompassing sources ranging from the Rig Veda and the Popol Vuh to the theology of Ptah and Dogon doctrines of word-based creation. These materials were examined through hermeneutic reading practices and comparatively evaluated using concept-oriented analytical categories. The findings indicate that cosmogonic myths operate beyond mere narrative description by structuring coherent models of creation in which cognitive intention and verbal articulation play constitutive roles. Full article

The style change detection (SCD) task involves identifying the locations of writing style changes in multi-authored documents. This task can be applied to plagiarism detection, security, and commerce applications. Introducing decoder-based Large Language Models (LLMs) marks a pivotal shift in applications. The segment boundaries for SCD models can be represented by concatenating two consecutive segments as pairs. However, LLMs usually restrict their input lengths, where the long-length inputs may exceed the restricted length. This paper seeks to bridge this gap and exploit the power of LLMs by introducing boundary-focused LLM Adaptation for SCD (BF-LLMA-SCD). The proposed solution adapts decoder-based LLMs for SCD using QLoRA. BF-LLMA-SCD truncates long-length input by preserving texts near an examined boundary while removing those at the other sides. BF-LLMA-SCD was trained on three PAN datasets. Comparison results with the top-performing SOTA solutions show that BF-LLMA-SCD achieved the best performance results in terms of F${}_1$ on PAN 2021 and PAN 2022/D1, while obtaining competitive results on PAN 2022/D3. BF-LLMA-SCD was also trained on an Arabic SCD dataset comprising three difficulty levels. It achieved an F${}_1$ score above 0.99 on easy instances. Full article

Autism-specific screening and developmental surveillance in primary care aid identification of autism. In this study, we assessed primary care clinicians’ (PCCs’) reported confidence in screening scores from the Modified Checklist for Autism in Toddlers, Revised (M-CHAT-R) and in their own diagnostic impressions. Four PCCs provided data for 50 children aged 16–36 months for whom they had any developmental concern. PCCs’ diagnostic impressions were “Definitely Autism” for 15 children (30%), “Unsure—Needs Further Evaluation” for 25 children (50%) and “Definitely Not Autism” for 10 children (20%). They reported High Confidence on the screener score in 33 cases (66%). Of the 17 cases for whom PCCs reported having Low Confidence on the M-CHAT-R, 14 children (82.3%) had a Low Likelihood score, with no significant association between M-CHAT-R likelihood and PCC’s confidence in the screening score. PCCs’ diagnostic impressions were concordant with the M-CHAT-R autism likelihood in 42% of cases, with a significantly higher mean in confidence rating when compared to the non-concordant cases. Language development and social engagement were the most frequently endorsed concerns by PCCs, with significant associations between these concerns and M-CHAT-R likelihood. Our results suggest that, when developmental concerns exist, PCCs may place greater confidence in the M-CHAT-R when scores indicate a higher likelihood of autism, and that confidence in their own diagnostic impressions may be associated with concordance with the screener score. Full article

This article explores Katherine Sonderegger’s thesis that in Christian prayer, not only does the person pray, but God prays. Though such an idea runs contrary to the settled conviction in Christian spirituality that the human person prays to God, this paper enquires into the idea that God also prays in the person with a study of Ignatius of Loyola’s Spiritual Diary. That record of his spiritual experiences suggests that not only did he listen to God’s prayer in him, but that this listening comprised a spiritual itinerary in which he was led into a deeper experience of God’s prayerful laboring in him. Following this itinerary, this article proceeds in three parts. First, a study of Ignatius’s prayer to the mediators reveals that in his petitions, he sought to hear the intercessory prayer of Mary and Jesus. Second, he found himself discovering a new way to name God as he celebrated the mass; that newness resided not in a new vocabulary but in his participation in the prayer of the Son to the Father. Finally, Ignatius experienced the grace of loqüela in which he heard a kind of celestial music whose tone and language moved him to a simple, contemplative admiration of God. More than the story of a mystic with an uncommon ability to listen to God, Ignatius’s journey into greater attention to God’s language within him is the story of grace, God’s life, which is always present, active, and audible in the believer’s prayer. Full article

Sheep wool is a keratin-based natural fiber increasingly explored as a low-impact reinforcement and multifunctional modifier in composites, enabling valorization of coarse or waste wool streams. This systematic review consolidates evidence on raw or minimally processed wool-reinforced composites across polymer matrices and mineral binders. Following a registered protocol and Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020, Scopus and Web of Science were searched for English-language journal articles (2015–2025), yielding 44 included studies after screening. Evidence mapping shows polymers dominate (33/44; thermosets 19/44), while mineral binders account for 11/44. Wool is mainly used as short fibers (27/44), with woven (9/44) and nonwoven/felt (8/44) architectures appearing in laminates and insulation products. Because heterogeneity limits pooled meta-analysis, outcomes are synthesized using matched-control comparisons where available (27/44) and interpreted with a TRiC appraisal (Transparency, Reproducibility, and Credibility). Mechanical effects are highly conditional: gains in impact/energy absorption and occasional tensile/flexural stress improvements coexist with frequent losses linked to dispersion, wetting/impregnation and void sensitivity. Functional trends are similarly system-dependent, with promising but uneven evidence for acoustic performance, variable thermal conductivity shifts, and formulation-driven fire behavior. Moisture uptake and durability emerge as principal translation bottlenecks, motivating minimum reporting and design practices to improve comparability and application readiness. Full article

Real-world Knowledge Graphs (KGs) are inherently incomplete, which hinders effective downstream reasoning. While Large Language Models (LLMs) possess powerful semantic capabilities, directly applying them to Knowledge Graph Completion (KGC) often leads to hallucinations and a lack of structural awareness. To address these challenges, we propose Embedding-Guided Instruction Tuning (EGIT), a novel framework that synergizes the structural precision of embedding models with the semantic reasoning of LLMs. Our approach operates in three key stages: (1) utilizing pre-trained embedding models to automatically synthesize high-quality, annotation-free instruction data; (2) fine-tuning the LLM with these structure-aware instructions to adapt it to the KGC task; and (3) employing a joint inference mechanism where the embedding model retrieves candidates and the fine-tuned LLM performs the final selection, thereby significantly reducing hallucinations. In extensive experiments, the best variant of EGIT achieves 7.0% and 2.5% improvements in Hits@1 on the FB15k-237 and WN18RR datasets, respectively. Full article

Manufacturing companies face significant challenges in leveraging artificial intelligence for equipment management due to high infrastructure costs and limited availability of labeled data for failures. While most manufacturing AI applications focus on structured sensor data, vast amounts of unstructured textual information containing valuable maintenance knowledge remain underutilized. This study presents a practical generative AI-based framework for structured information extraction that automatically converts unstructured equipment maintenance texts into predefined semantic fields to support predictive maintenance in manufacturing environments. We adopted and evaluated three representative generative models—Bidirectional and Auto-Regressive Transformers (BART) with KoBART, Text-to-Text Transfer Transformer (T5) with pko-t5-base, and the large language model Qwen—to generate structured outputs by extracting three predefined fields: failed components, failure types, and corrective actions. The framework enables the structuring of equipment management text data from Manufacturing Execution Systems (MES) to build predictive maintenance support systems. We validated the approach using a large-scale MES dataset consisting of 29,736 equipment maintenance records from a major automotive parts manufacturer, from which curated subsets were used for model training and evaluation. Our methodology employs Generative Pre-trained Transformer 4 (GPT-4) for initial dataset construction, followed by domain expert validation to ensure data quality. The trained models achieved promising performance when evaluated using extraction-aligned metrics, including exact match (EM) and token-level precision, recall, and F1-score, which directly assess field-level extraction correctness. ROUGE scores are additionally reported as a supplementary indicator of lexical overlap. Among the evaluated models, Qwen consistently outperformed BART and T5 across all extracted fields. The structured outputs are further processed through domain-specific dictionaries and regular expressions to create a comprehensive analytical database supporting predictive maintenance strategies. We implemented a web-based analytics platform enabling time-series analysis, correlation analysis, frequency analysis, and anomaly detection for equipment maintenance optimization. The proposed system converts tacit knowledge embedded in maintenance texts into explicit, actionable insights without requiring additional sensor installations or infrastructure investments. This research contributes to the manufacturing AI field by demonstrating a comprehensive application of generative language models to equipment maintenance text analysis, providing a cost-effective approach for digital transformation in manufacturing environments. The framework’s scalability and cloud-based deployment model present significant opportunities for widespread adoption in the manufacturing sector, supporting the transition from reactive to predictive maintenance strategies. Full article

Background: Methamphetamine (MA) use in people living with HIV (PLWH) has been linked to neurocognitive and behavioral dysregulation. We hypothesized that PLWH with active MA use (MAHIV) would show poorer cognitive performance, greater emotional and sleep burden, higher behavioral risk, and alterations in circulating biomarkers of immune activation and neuronal injury, relative to PLWH without MA use and HIV-negative Controls. Methods: Cross-sectional analytic study of 121 adults: PLWH with MA use (MAHIV, n = 40), PLWH without use (n = 42), and HIV-negative Controls (n = 39). Outcomes were ART discontinuation, physical activity, neurocognition (MoCA), depression (BDI), anxiety (GAD-7), sleep (PSQI), and substance use (ASSIST). Circulating biomarkers measured by ELISA: sCD14, neuron-specific enolase (NSE), S100B, and neurofilament light chain (NfL). Results: MAHIV participants had more frequent ART discontinuation than PLWH and the lowest physical activity. Chemsex with polysubstance use, condomless sex, and multiple partners were most prevalent in MAHIV. This group showed the highest anxiety and depressive burdens, and the greatest sleep disturbances. Global cognition (MoCA) was lowest in MAHIV, with significant deficits in executive function, memory, attention, and language; 82.5% had at least mild cognitive impairment. sCD14 was significantly higher in MAHIV than in PLWH and Controls, and NSE was elevated in both MAHIV and PLWH versus Controls. sCD14 correlated inversely with MoCA and positively with GAD-7 and BDI-II. Conclusions: Among PLWH, MA use is associated with greater ART nonadherence, syndemic mental-health and sleep disturbances, broader neurocognitive deficits, and elevations in circulating sCD14 and NSE. The sCD14–cognition and sCD14–mood relationships highlight chronic immune activation as a candidate pathway for neurocognitive and affective impairment and support sCD14 and NSE as potential stratification and monitoring biomarkers in MA-using PLWH. Full article