Search Results (76)

The global expansion of Korean language education has created an urgent need for scalable, objective, and consistent methods for assessing the writing skills of non-native (L2) learners. Traditional manual grading is resource-intensive and prone to subjectivity, while existing Automated Essay Scoring (AES) systems often struggle with the linguistic nuances of Korean and the specific error patterns of L2 writers. This paper introduces a novel hybrid AES system designed specifically for Korean L2 writing. The system integrates two complementary feature sets: (1) a comprehensive suite of conventional linguistic features capturing lexical diversity, syntactic complexity, and readability to assess writing form and (2) a novel semantic relevance feature that evaluates writing content. This semantic feature is derived by calculating the cosine similarity between a student’s essay and an ideal, high-proficiency reference answer generated by a Large Language Model (LLM). Various machine learning models are trained on the Korean Language Learner Corpus from the National Institute of the Korean Language to predict a holistic score on the 6-level Test of Proficiency in Korean (TOPIK) scale. The proposed hybrid system demonstrates superior performance compared to baseline models that rely on either linguistic or semantic features alone. The integration of the LLM-based semantic feature provides a significant improvement in scoring accuracy, more closely aligning the automated assessment with human expert judgments. By systematically combining measures of linguistic form and semantic content, this hybrid approach provides a more holistic and accurate assessment of Korean L2 writing proficiency. The system represents a practical and effective tool for supporting large-scale language education and assessment, aligning with the need for advanced AI-driven educational technology systems. Full article

As an essential medium for the cultural narrative of architectural space, studying the cognitive transformation mechanisms of traditional ethnic decorative patterns is critical for their effective preservation and innovative application. This research focuses on typical decorative motifs found in She ethnic architectural heritage, systematically classifying them into five categories—animal, plant, human figure, totem, and geometric—based on symbolic themes, formal structure, and cultural function. Correspondingly, 20 sets of standardized black-and-white line drawing stimuli were developed for experimental use. Methodologically, this study utilized the EyeLink 1000 eye-tracking system to acquire real-time gaze metrics, including fixation duration and saccadic amplitude, as well as pupil dilation responses from participants engaged in a controlled pattern observation task. Immediately after observation, participants completed a semantic differential assessment using a five-point Likert scale. Data analysis employed descriptive statistics, analysis of variance (ANOVA), Kruskal–Wallis tests, and Bonferroni-adjusted post hoc comparisons (α = 0.05). Attention allocation was further examined through heatmaps and gaze trajectory visualizations to provide comprehensive insight into visual engagement. Two principal findings were identified: first, male participants showed a predominant focus on holistic structural composition and cultural symbol representation, whereas female participants exhibited a processing bias towards fine details; second, concrete symbols imbued with historical significance elicited more pronounced emotional responses, while abstract geometric patterns necessitated formal reconstruction to enhance cognitive accessibility. These findings offer empirical support for gender-inclusive architectural design strategies and inform practical approaches for safeguarding cultural heritage within contemporary architectural environments. Consequently, modern reinterpretation of traditional decorative patterns should balance cultural narrative fidelity with functional adaptation, achieving inclusive expression through contextual reconstruction and interactive design strategies. Future research directions include expanding participant demographics to encompass cross-cultural cohorts and incorporating multimodal neuroimaging techniques to elucidate the underlying cognitive and affective mechanisms, thereby advancing the sustainable transmission and innovation of ethnic cultural heritage. Full article

Identifying homologous vulnerabilities across diverse IoT firmware images is critical for large-scale vulnerability auditing and risk assessment. However, existing approaches often rely on coarse-grained components or single-dimensional metrics, lacking the semantic granularity needed to capture cross-firmware vulnerability relationships. To address this gap, we propose FirmVulLinker, a semantic profiling framework that holistically models firmware images across five dimensions: unpacking signature sequences, filesystem semantics, interface exposure, boundary binary symbols, and sensitive parameter call chains. These multi-dimensional profiles enable interpretable similarity analysis without requiring prior vulnerability labels. We construct an evaluation dataset comprising 54 Known Defective Firmware (KDF) images with 74 verified vulnerabilities and assess FirmVulLinker across multiple correlation tasks. Compared to state-of-the-art techniques, FirmVulLinker achieves higher precision with substantially lower false-positive and false-negative rates. Notably, it identifies and reproduces 53 previously undisclosed N-day vulnerabilities in firmware images not listed as affected at the time of public disclosure, effectively extending the known impact scope. Our results demonstrate that FirmVulLinker enables scalable, high-fidelity homologous vulnerability analysis, offering a new perspective on understanding cross-firmware vulnerability patterns in the IoT ecosystem. Full article

Occlusion remains a major challenge in person re-identification, as it often leads to incomplete or misleading visual cues. To address this issue, we propose a dual-branch occlusion-aware network (DOAN), which explicitly and implicitly enhances the model’s capability to perceive and handle occlusions. The proposed DOAN framework comprises two synergistic branches. In the first branch, we introduce an Occlusion-Aware Semantic Attention (OASA) module to extract semantic part features, incorporating a parallel channel and spatial attention (PCSA) block to precisely distinguish between pedestrian body regions and occlusion noise. We also generate occlusion-aware parsing labels by combining external human parsing annotations with occluder masks, providing structural supervision to guide the model in focusing on visible regions. In the second branch, we develop an occlusion-aware recovery (OAR) module that reconstructs occluded pedestrians to their original, unoccluded form, enabling the model to recover missing semantic information and enhance occlusion robustness. Extensive experiments on occluded, partial, and holistic benchmark datasets demonstrate that DOAN consistently outperforms existing state-of-the-art methods. Full article

To address the widespread challenges of significant multi-category dental morphological variations and interference from overlapping anatomical structures in panoramic dental X-ray images, this paper proposes a dual-stream dental segmentation model based on Transformer heterogeneous feature complementarity. Firstly, we construct a parallel architecture comprising a Transformer semantic parsing branch and a Convolutional Neural Network (CNN) detail capturing pathway, achieving collaborative optimization of global context modeling and local feature extraction. Furthermore, a Pooling-Cooperative Convolutional Module was designed, which enhances the model’s capability in detail extraction and boundary localization through weighted centroid features of dental structures and a latent edge extraction module. Finally, a Semantic Transformation Module and Interactive Fusion Module are constructed. The Semantic Transformation Module converts geometric detail features extracted from the CNN branch into high-order semantic representations compatible with Transformer sequential processing paradigms, while the Interactive Fusion Module applies attention mechanisms to progressively fuse dual-stream features, thereby enhancing the model’s capability in holistic dental feature extraction. Experimental results demonstrate that the proposed method achieves an IoU of 91.49% and a Dice coefficient of 94.54%, outperforming current segmentation methods across multiple evaluation metrics. Full article

Understanding how people perceive urban streetscapes is essential for enhancing the visual quality of the urban environment and optimizing street space design. While perceptions are shaped by the interplay of multiple visual elements, existing studies often isolate single semantic features, overlooking their combinations. This study proposes a Landscape Element Combination Extraction Method (SLECEM), which integrates the UniSal saliency detection model and semantic segmentation to identify landscape combinations that play a dominant role in human perceptions of streetscapes. Using street view images (SVIs) from the central area of Futian District, Shenzhen, China, we further construct a multi-dimensional feature–perception coupling analysis framework. The key findings are as follows: 1. Both low-level visual features (e.g., color, contrast, fractal dimension) and high-level semantic features (e.g., tree, sky, and building proportions) significantly influence streetscape perceptions, with strong nonlinear effects from the latter. 2. K-Means clustering of salient landscape element combinations reveals six distinct streetscape types and perception patterns. 3. Combinations of landscape features better reflect holistic human perception than single variables. 4. Tailored urban design strategies are proposed for different streetscape perception goals (e.g., beauty, safety, and liveliness). Overall, this study deepens the understanding of streetscape perception mechanisms and proposes a highly operational quantitative framework, offering systematic theoretical guidance and methodological tools to enhance the responsiveness and sustainability of urban streetscapes. Full article

With the rapid development of digital technologies, data-driven methods have demonstrated commendable performance in the toxic text detection task. However, several challenges remain unresolved, including the inability to fully capture the nuanced semantic information embedded in text languages, the lack of robust mechanisms to handle the inherent uncertainty of text languages, and the utilization of static fusion strategies for multi-view information. To address these issues, this paper proposes a comprehensive and dynamic toxic text detection method. Specifically, we design a multi-view feature augmentation module by combining bidirectional long short-term memory and BERT as a dual-stream framework. This module captures a more holistic representation of semantic information by learning both local and global features of texts. Next, we introduce an entropy-oriented invariant learning module by minimizing the conditional entropy between view-specific representations to align consistent information, thereby enhancing the representation generalization. Meanwhile, we devise a trustworthy text recognition module by defining the Dirichlet function to model uncertainty estimation of text prediction. And then, we perform the evidence-based information fusion strategy to dynamically aggregate decision information between views with the help of the Dirichlet distribution. Through these components, the proposed method aims to overcome the limitations of traditional methods and provide a more accurate and reliable solution for toxic language detection. Finally, extensive experiments on the two real-world datasets show the effectiveness and superiority of the proposed method in comparison with seven methods. Full article

Current tourism route recommendation systems often overemphasize popular destinations, thereby overlooking geographical accessibility between attractions and the experiential coherence of the journey. Leveraging multidimensional attribute perceptions derived from tourist reviews, this study proposes a Spatial–Semantic Integrated Model for Tourist Attraction Representation (SSIM-TAR), which holistically encodes the composite attributes and multifaceted evaluations of attractions. Integrating these multidimensional features with inter-attraction relationships, three relational metrics are defined and fused: spatial proximity, resonance correlation, and thematic-sentiment similarity, forming a Tourist Attraction Multidimensional Association Network (MAN-SRT). This network enables precise characterization of complex inter-attraction dependencies. Building upon MAN-SRT, the Tourism Sentiment Chain (TSC) model is proposed that incorporates geographical accessibility, associative resonance, and thematic-sentiment synergy to optimize the selection and sequential arrangement of attractions in personalized route planning. Results demonstrate that SSIM-TAR effectively captures the integrated attributes and experiential quality of tourist attractions, while MAN-SRT reveals distinct multidimensional association patterns. Compared with popular platforms such as “Qunar” and “Mafengwo”, the TSC approach yields routes with enhanced spatial efficiency and thematic-sentiment coherence. This study advances tourism route modeling by jointly analyzing multidimensional experiential quality through spatial–semantic feature fusion and by achieving an integrated optimization of geographical accessibility and experiential coherence in route design. Full article

The Netherlands has established climate-adaptive strategies shaped by its long history of water-related climate events, such as the floods in 1421 and 1953. UNESCO World Heritage (WH) properties in The Netherlands reflect centuries of human intervention and natural processes to adapt and mitigate climate challenges, including spatial design and hydraulic engineering. The Dutch Climate Research Initiative also highlights cultural heritage as an integral component in preparing for the 2026 National Climate Adaptation Strategy. This article aims to unveil climate-adaptive World Heritage management strategies (CAWHMSs), using WH properties in The Netherlands as a case study. It collects textual data from Statements of Outstanding Universal Value, State of Conservation Reports by the State Parties and management plans. Through qualitative coding and keywords aggregation of the documents, the visualised results of a Sankey diagram and two semantic networks confirmed two CAWHMSs: conservation and developing WH properties as collaborative knowledge hubs. Conservation supports regulating urban climate and sustainable water management. As collaborative knowledge hubs, multidisciplinary sectors explore opportunities to align WH properties with broader sustainable development initiatives. They also deepen younger generations’ awareness of cultural and natural significance relevant to mitigating climate threats. The results emphasise WH as a contributor to climate adaptation. Cross-sectoral stakeholders can advance holistic climate adaptation efforts using CAWHMSs. Full article

Malware continues to evolve rapidly, posing significant challenges to network security. Traditional signature-based detection methods often struggle to cope with advanced evasion techniques such as polymorphism, metamorphism, encryption, and stealth, which are commonly employed by cybercriminals. As a result, these conventional approaches frequently fail to detect newly emerging malware variants in a timely manner. To address this limitation, Zero-Shot Learning (ZSL) has emerged as a promising alternative, offering improved classification capabilities for previously unseen malware samples. ZSL models leverage auxiliary semantic information and binary feature representations to enhance the recognition of novel threats. This study proposes a Transductive Zero-Shot Learning (TZSL) model based on the Vector Quantized Variational Autoencoder (VQ-VAE) architecture, integrated with a malware knowledge graph constructed from sandbox behavioral analysis of ransomware families. The model is further optimized through hyperparameter tuning to maximize classification performance. Evaluation metrics include per-family classification accuracy, precision, recall, F1-score, and Receiver Operating Characteristic (ROC) curves to ensure robust and reliable detection outcomes. In particular, the harmonic mean (H-mean) metric from the Generalized Zero-Shot Learning (GZSL) framework is introduced to jointly evaluate the model’s performance on both seen and unseen classes, offering a more holistic view of its generalization ability. The experimental results demonstrate that the proposed VQ-VAE model achieves an F1-score of 93.5% in ransomware classification, significantly outperforming other baseline models such as LeNet-5 (65.6%), ResNet-50 (71.8%), VGG-16 (74.3%), and AlexNet (65.3%). These findings highlight the superior capability of the VQ-VAE-based TZSL approach in detecting novel malware variants, improving detection accuracy while reducing false positives. Full article

This article presents a comprehensive review of short text clustering using state-of-the-art methods: Bidirectional Encoder Representations from Transformers (BERT), Term Frequency-Inverse Document Frequency (TF-IDF), and the novel hybrid method Latent Dirichlet Allocation + BERT + Autoencoder (LDA + BERT + AE). The article begins by outlining the theoretical foundation of each technique and its merits and limitations. BERT is critiqued for its ability to understand word dependence in text, while TF-IDF is lauded for its applicability in terms of importance assessment. The experimental section compares the efficacy of these methods in clustering short texts, with a specific focus on the hybrid LDA + BERT + AE approach. A detailed examination of the LDA-BERT model’s training and validation loss over 200 epochs shows that the loss values start above 1.2 and quickly decrease to around 0.8 within the first 25 epochs, eventually stabilizing at approximately 0.4. The close alignment of these curves suggests the model’s practical learning and generalization capabilities, with minimal overfitting. The study demonstrates that the hybrid LDA + BERT + AE method significantly enhances text clustering quality compared to individual methods. Based on the findings, the study recommends the optimum choice and use of clustering methods for different short texts and natural language processing operations. The applications of these methods in industrial and educational settings, where successful text handling and categorization are critical, are also addressed. The study ends by emphasizing the importance of the holistic handling of short texts for deeper semantic comprehension and effective information retrieval. Full article

Burn injuries lead to significant physical and psychological consequences, including chronic pain, post-traumatic stress, depression, and social isolation. Mindfulness-based interventions (MBIs) have been proposed as a holistic approach to address these challenges in burn rehabilitation. This systematic review evaluates the efficacy of dispositional mindfulness and MBIs, including mindfulness meditation, yoga, and self-compassion training, in managing pain, emotional distress, and psychosocial adaptation in burn survivors. A comprehensive literature search was conducted through MEDLINE and Web of Science, covering studies up to February 2025, with additional papers retrieved from Google Scholar and Semantic Scholar. Studies were included if they reported quantitative data on the effects of MBIs in burn patients and/or their families, excluding opinion pieces, editorials, reviews, and qualitative studies. After screening 91 studies retrieved from the databases and adding a compelling paper retrieved from the other sources explored, 12 studies were included in the final pool, categorized into cross-sectional studies (n = 6), and intervention studies (n = 6). The extracted data included publication year, research design, sample characteristics, intervention details, main findings, and data for quality assessment. The synthesis of the results suggests that mindfulness is associated with reduced psychological symptoms, improved emotional regulation, and enhanced self-compassion, leading to better coping strategies and social reintegration. However, the long-term efficacy of MBIs remains inconclusive, and further research is needed to differentiate mindfulness-specific effects from those of general physical exercise. Evidence also suggests that mindfulness interventions may reduce anxiety and secondary trauma in children with burns and their caregivers. This review highlights the potential of MBIs as adjuncts to conventional burn rehabilitation programs, but further high-quality trials are needed to establish their sustained efficacy and to understand the specific benefits of mindfulness. Full article

Contrastive image representation learning has been essential for pre-training vision foundation models to deliver excellent transfer learning performance. It was originally developed based on instance discrimination, which focuses on instance-level recognition tasks. Lately, the focus has shifted to directly working on the dense spatial features to improve transfer performance on dense prediction tasks such as object detection and semantic segmentation, for which pixel-level and region-level contrastive learning methods have been proposed. Region-level methods usually employ region-mining algorithms to capture holistic regional semantics and address the issue of semantically inconsistent scene image crops, as they assume that pixel-level learning struggles with both. In this paper, we revisit pixel-level learning’s potential and show that (1) it can effectively and more efficiently learn holistic regional semantics and (2) it intrinsically provides tools to mitigate the impact of semantically inconsistent views involved with scene-level training images. We prove this by proposing PixCon, a pixel-level contrastive learning framework, and testing different positive matching strategies based on this framework to rediscover the potential of pixel-level learning. Additionally, we propose a novel semantic reweighting approach tailored for pixel-level learning-based scene image pre-training, which outperforms or matches the performance of previous region-level methods in object detection and semantic segmentation tasks across multiple benchmarks. Full article

In an era of evolving scholarly ecosystems, machine learning (ML) and artificial intelligence (AI) have become pivotal in advancing research impact analysis. Despite their transformative potential, the fragmented body of literature in this domain necessitates consolidation to provide a comprehensive understanding of their applications in multidimensional impact assessment. This study bridges this gap by employing bibliometric methodologies, including co-authorship analysis, citation burst detection, and advanced topic modelling using BERTopic, to analyse a curated corpus of 1608 scholarly articles. Guided by three core research questions, this study investigates how ML and AI enhance research impact evaluation, identifies dominant methodologies, and outlines future research directions. The findings underscore the transformative potential of ML and AI to augment traditional bibliometric indicators by uncovering latent patterns in collaboration networks, institutional influence, and knowledge dissemination. In particular, the scalability and semantic depth of BERTopic in thematic extraction, combined with the visualisation capabilities of tools such as CiteSpace and VOSviewer, provide novel insights into the dynamic interplay of scholarly contributions across dimensions. Theoretically, this research extends the scientometric discourse by integrating advanced computational techniques and reconfiguring established paradigms for assessing research contributions. Practically, it provides actionable insights for researchers, institutions, and policymakers, enabling enhanced strategic decision-making and visibility of impactful research. By proposing a robust, data-driven framework, this study lays the groundwork for holistic and equitable research impact evaluation, addressing its academic, societal, and economic dimensions. Full article

Safety inspection of building façades in urban settings is critical for public safety, as many incidents/accidents frequently occur due to falls from façades. This inspection is required for condition assessment of current states and their comparison to assessments conducted in a previous inspection cycle. The current practice of façade condition assessment relies on static, scattered, and textual depictions, which prevents inspectors from having a comprehensive view of façade condition and comparing it to the previous inspection cycle’s findings. Integrated visualization of spatial and semantic data and providing this data based on the preferences of the decision makers are proven to be effective in various decision domains. This study builds on previous research efforts on integrated visualization techniques to identify façade inspectors’ preferences in comprehending inspection findings. Through the design of low-fidelity visualization prototypes, this study first identifies highly frequently preferred visualization techniques by inspectors. Based on these, this study then quantifies the impact of these identified visualization preferences on accuracy and efficiency of decisions in relation to condition assessment of building façades though a set of high-fidelity prototypes and user studies. The results show that integrated visualization of façade conditions has the potential to bring the efficiency of capturing a holistic view of the façade conditions up to 65% and increase the accuracy of decisions up to 41%. The findings of this study directly benefit inspection companies and city agencies who can deploy visualization techniques that are impactful for façade condition assessment in order to holistically assess building façades. Full article