Search Results (181)

Historic buildings often exhibit high energy use intensity (EUI), while conservation constraints limit envelope retrofits, making it difficult to identify robust and actionable operational predictors. Using four in-use historic buildings in Shenyang, China, this study presents a pilot methodological demonstration with a controlled-comparability workflow consisting of two linked layers: (i) a Driver layer of intervenable operational variables and (ii) a Pathway layer of calibrated EnergyPlus heat-balance terms for physics-informed interpretation. Three importance approaches (Spearman, wrapper RFE with XGBoost, and Random Forest) are compared; rankings are fused via reciprocal rank fusion, and stability is tested using cross-period rolling validation across Top-K feature sets. After similarity screening, EUI variation is better explained by operational predictors and the corresponding simulated loss channels than by macro-scale structural heterogeneity. Infiltration-related indicators and envelope/infiltration loss components remain consistently prominent, while Spearman importance is less stable in the Pathway layer under seasonal switching and nonlinear coupling. A Top-10 subset provides a favorable accuracy–stability trade-off. The proposed Driver–Pathway mapping supports conservation-compatible prioritization hypotheses within a simulation-consistent interpretive framework; findings are associational and context dependent and should be validated through field measurements and experimental or quasi-experimental studies before prescriptive claims are made. Full article

To address the challenges of modal differences in multimodal farmland images and insufficient segmentation accuracy for small targets, this paper proposes a multi-source image fusion branch (DAMFusion) based on modal competitive selection. The branch dynamically selects infrared and visible light features through the Competitive Query Module (CQM) using Top-K screening, combined with IOU-aware loss optimization to avoid cross-modal interference. The multimodal fusion module (MMFormer) employs cross-modal attention and symmetric mechanisms, enhancing single-modal features through a self-enhancement module and unifying multimodal distributions via linear projection. The Boundary Region Attention Multi-level Fusion Module (BRM) extracts boundary information through feature differencing, strengthens it with spatial attention, and fuses it with shallow features to achieve cross-layer detail recovery. Through the collaborative design of dynamic modal feature selection, cross-modal distribution unification, and boundary region enhancement, DAMFusion effectively solves the problems of multimodal differences and small target segmentation in multispectral images, providing precise feature representation for fine farmland segmentation. Experiments on the OUC-UAV-MSEG dataset show that DAMFusion achieves 93.25% OA, 91.71% F1, and 89.70% mIoU, demonstrating clear advantages over representative comparison methods. In addition, ablation results verify the effectiveness of the proposed modules, where CQM improves OA from 91.00% to 93.25%, confirming the importance of discriminative modality selection before fusion. Full article

With the continuous emergence of exam question types, accurate classification of knowledge points is crucial for intelligent exam analysis. Existing methods focus on text or text–image fusion but largely ignore spatial layout. To address this limitation, we propose a heterogeneous layout-aware cross-modal framework for knowledge point classification. The architecture begins with an encoding module where independent text and layout encoders extract semantic content and spatial configurations, respectively. We then design a layout-aware enhancing module consisting of two parallel cross-modal blocks, namely a Layout-Aware Text-Enhancing block and a Context-Aware Layout-Enhancing block. This module supports the bidirectional fusion of text and layout features and generates a comprehensive representation that integrates both semantic and spatial information. Furthermore, a dynamic router with top-k expert selection is introduced to dynamically adapt to question-specific knowledge distributions and focus on core knowledge points for precise classification. Experimental results demonstrate that our method effectively integrates text and layout information, significantly enhancing performance on the proposed QType-EDU dataset. The approach achieves 91.56% accuracy for coarse-grained classification and 80.58% for fine-grained classification, with an overall F1-score of 91.39%, surpassing all baseline models. Full article

Background/Objectives: This study aims to develop a comprehensive Clinical Decision Support System (CDSS) that integrates multi-sequence sacroiliac joint (SIJ) MRIs with rheumatological, clinical, and laboratory findings into the decision-making process for the early diagnosis of axial spondyloarthritis (axSpA), incorporating segmentation-supported explainability. Methods: Multi-sequence SIJ MRI data (T1-WI, T2-WI, STIR, and PD-WI) were analysed from 367 participants (n = 193 axSpA; n = 174 non-axSpA controls). Sequence-based classification was performed using VGG16, ResNet50, DenseNet121, and InceptionV3 models; additionally, a lightweight and parameter-efficient SacroNet architecture was developed. Slice-level probability scores were converted to patient-level scores using the Dynamic Top-K Averaging method. Image-based scores were combined with a logistic regression-based clinical risk score using weighted linear integration (0.60 image/0.40 clinical) and a conservative threshold (τ = 0.70). Grad-CAM was applied for visual interpretability. Furthermore, to support the diagnostic outcomes with precise spatial data, active inflammation in STIR and T2-WI sequences was segmented. For this purpose, the MDC-UNet model was employed and compared with baseline U-Net derivatives. Results: Sequence-specific analysis showed VGG16 performing best on T1-WI (AUC = 0.920; Accuracy = 0.878) and DenseNet121 on STIR (AUC = 0.793; Accuracy = 0.771). The SacroNet architecture provided competitive classification performance at the patient level despite its low number of parameters (~110 K). Furthermore, MDC-UNet successfully segmented active inflammation, yielding Dice scores of 0.752 (HD95: 19.25) for STIR and 0.682 (HD95: 26.21) for T2-WI. Conclusions: The findings demonstrate that patient-level decision integration based on multi-sequence MRI, when used in conjunction with clinical risk scoring and segmentation-assisted interpretability, can provide a feasible and interpretable DSS framework for the early diagnosis of axSpA. Full article

Generative AI systems increasingly mediate how students retrieve literature and generate citations, shifting methodological rigor toward the maintenance of an auditable evidence trail. This study audits the search stage of AI-assisted literature review work, focusing on retrieval performance and citation traceability rather than downstream screening or synthesis. Four widely accessible tools were compared across two retrieval postures, and Boolean queries were executed against Scopus and evaluated against a DOI-verified librarian baseline built from Scopus, Web of Science, and Google Scholar. Using a canonical prompt and a bounded top-k capture rule (k = 20), each bibliographic record was evaluated for DOI traceability, DOI resolution integrity, metadata accuracy, and run-to-run drift. Records were screened through staged title/abstract and full-text eligibility review, and the final set included 37 studies after quality appraisal was 37 studies. Across sixteen audit runs, natural-language prompting frequently produced under-target yields, recurrent integrity failures, and low overlap with the librarian benchmark. Boolean translation improved run completion and increased the proportion of auditable records, but reproducibility remained unstable across repeated runs. These findings show that correctness at the record level does not ensure stability at the evidence-set level. Limitations include the bounded tool set, the search-stage focus, and the absence of downstream screening or synthesis evaluation. Retrieval posture, therefore, emerges as a practical governance lever for AI-assisted literature review workflows and supports the use of a student-facing verification checklist anchored in DOI verification and transparent protocol capture. This research received no external funding. OSF registration: Open Science Framework, 10.17605/OSF.IO/U8NHT. The manuscript reports the final included set as n = 37, states no external funding, and lists the OSF registration DOI. Full article

Large language models (LLMs) and diffusion-based image generators can rapidly produce architectural ideas and imagery, yet translating conceptual narratives into massing composition is often implicit and difficult to reproduce. In this paper, we present a knowledge-augmented two-stage workflow for architectural concept-to-massing generation and evaluation. The outputs are represented as axonometric massing proxy images, which serve as 2D visual proxies for early-stage massing refinement rather than editable 3D models. The workflow integrates a prototype library and Knowledge Graph (KG) routing to map narrative cues into executable strategy and operation tokens and compile stage-specific prompts. Stage 1 produces structural concept sketches emphasizing legible composition, while Stage 2 generates axonometric massing proxy images conditioned on Stage 1 sketches to stabilize composition across candidates. Under a fixed sampling budget, candidates are ranked using a rubric-based scoring protocol with Top-K selection, and evaluation signals can be written back to update prompt compilation iteratively. Across diverse project briefs, ablation studies demonstrate that knowledge augmentation improves constraint compliance and composition readability while maintaining controlled diversity for early exploration. We report expert ratings together with paired statistical tests to support reproducible comparisons. Full article

Benggang (Benggang), a typical landform characterized by severe erosion and a geohazard in the red-soil hilly regions of southern China, is characterized by a fragmented texture, irregular boundaries, and high similarity to background objects such as bare soil and roads, which poses a dual challenge of “multiscale variability + strong noise” for automated identification at regional scales. To address insufficient information from a single modality and the limited representation of cross-scale features, this study proposes a dual-stream feature-fusion network (DF-Net) for multisource data consisting of a digital orthophoto map (DOM) and a digital elevation model (DEM). The method adopts ResNeSt50d as the backbone of the two branches: on the DOM side, a Canny-edge channel is stacked to enhance high-frequency boundary information; on the DEM side, derived terrain factors, including slope, aspect, curvature, and hillshade, are introduced to provide morphological constraints. In the cross-modal fusion stage, a multiscale sparse attention fusion module is designed, which acquires contextual information via multiwindow average pooling and suppresses noise interference through top-K sparsification. In the decision stage, a multibranch ensemble is employed to improve classification stability. Taking Anxi County, Fujian Province, as the study area, a coregistered dataset of GF-2 (1 m) DOM and ALOS (12.5 m) DEMs is constructed, and a zonal partitioning strategy is adopted to evaluate the model’s generalization ability. The experimental results show that DF-Net achieves 97.44% accuracy, 85.71% recall, and an 82.98% F1 score in the independent test zone, outperforming multiple mainstream CNN/transformer classification models. This study indicates that the strategy of “multimodal feature enhancement + sparse attention fusion” tailored to Benggang erosional landforms can significantly improve recognition performance under complex backgrounds, providing technical support for rapid Benggang surveys and governance-effectiveness assessments. Full article

Many real-world screening tasks in venture capital must rank large start-up candidate pools under conditions of tight review capacity, time-varying information, and rare investment success outcomes. When datasets are constructed retrospectively, post-decision updates can leak into features and inflate performance, especially with random splits. This study proposes a leakage-aware, time-based evaluation framework for capacity-constrained screening formulated as a top-K ranking problem. Using a dataset of 117,141 early-stage firms as an empirical testbed, features were constructed strictly as of a reference time $t_0$, a 180-day temporal embargo was enforced around the train–test boundary, and generalization was assessed with time-ordered splits. Because venture capital decisions are made on a shortlist, evaluation emphasizes ranking quality using PR-AUC, Lift@K, Precision@K/Recall@K, and NDCG@K, reported with bootstrap confidence intervals. Under this leakage-aware protocol and with strong class imbalance, maturity-related signals achieve the strongest PR-AUC (0.0144), while team and combined signals yield the best top-50 shortlist concentration. Finally, probability calibration substantially improves reliability for threshold planning (Brier score reduced from 0.0972 to 0.0161 with sigmoid calibration) while leaving ranking essentially unchanged. Overall, the study provides a leakage-aware evaluation template and an interpretable baseline for time-dependent venture capital screening tasks involving start-up selection, investment success prediction, leakage risk, and limited review capacity. Full article

The rapid growth of digital tourism platforms has intensified information overload and decision complexity for both locals and travelers, while operators struggle to differentiate their offerings and sustain profitable, data-driven e-commerce models. This paper presents Doroob, a big data and artificial intelligence (AI)-driven, context-aware recommendation system that integrates traditional recommender techniques with real-time facial emotion recognition (FER) to enable intelligent tourism commerce. Doroob combines three AI-based recommendation strategies: smart adaptive recommendation (SAR) collaborative filtering, a Vowpal Wabbit-based context-aware model, and a LightFM hybrid model. It trained on datasets built from the Google Places API and enriched with ratings adapted from MovieLens. FER, implemented with DeepFace and OpenCV, analyzes short video segments as users browse destination details, converts emotion scores into 1–5 satisfaction ratings, and stores this implicit feedback alongside explicit ratings to support adaptive, emotion-aware personalization. Experimental results show that the context-aware model achieves the strongest top-K ranking performance, the hybrid LightFM model yields the highest AUC of 0.95, and the SAR model provides the most accurate rating predictions, demonstrating that combining contextual modeling and FER-based implicit feedback can enhance personalization, mitigate cold-start, and support data-driven promotion of local tourist services in intelligent e-commerce ecosystems. Full article

Predicting the future impact of technical economic documents at release time is challenging due to delayed supervision signals, long-tailed label distributions, and time- and domain-dependent shifts in language and topics. Moreover, similarity graphs derived from text embeddings can be noisy due to boilerplate and evolve under temporal drift, making robustness and leakage-free evaluation essential. We formulate grant-time patent impact prediction as a node classification and within-domain ranking problem on a large-scale semantic similarity document graph built from patent text embeddings, avoiding any future citation leakage. The document graph is constructed via ANN Top-K retrieval and similarity thresholding, enabling scalable and reproducible sparsification on hundreds of thousands of nodes. We propose GraphGPT-Patent, which adapts a reversible graph-to-sequence foundation backbone to local subgraphs extracted from the similarity network. The model incorporates time- and domain-conditioned edge reliability to suppress drift-induced and template-driven pseudo-similarity, and optimizes a joint objective coupling high-impact classification with ranking consistency within comparable groups. Experiments on USPTO granted patents (2000–2022) across three high-volume CPC domains and three evaluation horizons show consistent gains over text-only and GNN baselines, achieving up to 0.94 recall for the positive class and improved macro-average recall across nine settings. Temporal shift analyses further quantify the effect of training-data freshness, while explanation subgraphs provide auditable structural evidence of model decisions. The proposed framework offers an effective graph-based learning pipeline for scalable impact prediction and downstream triage under strict information constraints. Full article

Small pest detection in ultra-high-resolution forestry images is challenging due to extreme scale variation, complex backgrounds, and limited annotated data. To address these issues, we propose SSFPDet (Semi-Supervised Forest Pest Detector), a semi-supervised object detection framework designed for low-annotation settings. Built upon the Soft Teacher paradigm, SSFPDet integrates a YOLO-T-based overlapping slicing strategy, a Top-K pseudo-label selection mechanism, and a Kullback–Leibler (KL) divergence-based distribution alignment constraint. The slicing strategy enhances small-object representation without modifying the detector backbone, while the Top-K and KL modules improve pseudo-label reliability and semantic consistency during training. Under the 20% labeled setting, SSFPDet achieves an mAP@0.5:0.95 of 46.6, outperforming the baseline by 0.7 points. Notably, small-object detection performance (AP_S) improves by 6.6 percentage points. Ablation studies confirm the complementary contributions of spatial slicing and semantic alignment. Overall, SSFPDet provides a practical and scalable solution for high-resolution forestry pest monitoring under limited supervision. Full article

Accessing and understanding medication information can be challenging for many people, especially when patient information leaflets (PILs) are long, complex, and printed in small font. This study presents MediChat, an open-source, locally executable chatbot designed to provide reliable, easy-to-read answers to medication-related questions based exclusively on official PILs. MediChat follows a retrieval-augmented generation (RAG) architecture: PILs from the Austrian Medicinal Product Index are received via API, converted to text, split into overlapping chunks, embedded, and stored in a Chroma vector database. From there the top-k relevant chunks are retrieved, and Llama 3.1 generates German responses based on this evidence. The system was evaluated using a hybrid framework. Quantitatively, 200 yes/no questions across ten drugs were answered with 80% accuracy, overall precision 0.977, recall 0.686, F1-score 0.806, and a mean response time of 727 ms. Qualitatively, two personas were used in eight simulated dialogues. Response times were around 1.1–1.3 s, and task completion exceeded 85% with high ratings for relevance and quantity. These results indicate that an open-source RAG chatbot can deliver leaflet-grounded, user-friendly medication information and provide a reproducible template for future healthcare chatbot evaluations. Full article

Infrared small target detection (IRSTD) is a fundamental task in remote sensing-based surveillance and early warning systems. However, extremely small target size, low signal-to-noise ratio, and complex background clutter make reliable detection highly challenging. To address these issues, we propose a Hierarchical Perception Fusion Network (HPFNet) for IRSTD. Specifically, the Patch-Wise Context Feature Extraction module (PCFE) jointly integrates the Patch Nonlocal Block, convolutional blocks and attention mechanism to enable global–local feature extraction and enhancement, thereby strengthening weak target representations. In addition, the Multi-Level Sparse Cross-Fusion module (MSCF) explicitly performs cross-level feature interaction between encoder and decoder representations, enabling effective fusion of low-level spatial details and high-level semantic cues. A dual Top-K sparsification mechanism is adopted to filters’ irrelevant background features, enabling the attention mechanism to focus more on the target region and thereby bolstering the discriminative power of feature representation. Finally, the Efficient Upsampling Module (EUM) combines upsampling with multi-branch dilated convolutions to enhance feature reconstruction and improve localization accuracy. Extensive experiments on publicly available benchmark datasets demonstrate that HPFNet consistently outperforms existing state-of-the-art IRSTD methods. Full article

Fifth generation (5G) networks have significantly enhanced connectivity, speed, and reliability, transforming industries with faster and more efficient communication. The Internet of Things (IoT) has introduced unprecedented convenience and automation, revolutionizing sectors such as healthcare, finance, and smart infrastructure. However, both 5G networks and IoT environments are experiencing a high frequency of attacks. Intrusion detection systems (IDSs) built on federated learning (FL) are being proposed to boost data privacy and security. However, these IDSs are related with the inherent drawbacks of FL, namely the existence of non-independently and identically (non-IID) distributed features and the machine learning model complexity. To address these limitations, we present a study that integrates a Mixture of Experts (MoE) into an FL setting in the task of intrusion detection. Specifically, to mitigate the issues of model complexity within the FL setting, we use a sparsely gated MoE layer consisting of a router/gating network and a set of experts. Only a subset of experts is selected via applying noisy top-k gating. To alleviate the issue of non-IID data, we adopt the Label-based Dirichlet Partition method, utilizing Dirichlet sampling with a hyperparameter $\alpha$ to simulate a label-based non-IID data distribution. Four FL strategies are employed. We perform our experiments on the 5G-NIDD and BoT-IoT datasets. Findings show that the proposed approach achieves competitive performance across both datasets under heterogeneous federated settings. Full article

Integrating flood-response datasets across municipalities is often hindered by heterogeneous and non-standard variable names, a challenge amplified in Korean by local naming conventions and linguistic variation. This study addresses scalable schema alignment to standardize municipal flood datasets with reduced manual effort while maintaining semantic consistency for downstream modeling. We propose a BERT-based schema matching framework that augments standardized attribute names with paraphrases generated by a generative language model and filtered to reduce semantic drift. Both standardized and target variable names are encoded using a flood-domain-adapted Korean BERT model, and candidate correspondences are retrieved via cosine-similarity ranking to produce top-k match suggestions for automated or human-in-the-loop alignment. Experiments on real flood-related tables from Busan and Incheon, evaluated jointly to diversify variable expressions, show that augmentation substantially improves top-k retrieval accuracy. In the combined evaluation, Hit@5 improves from 0.71 to 0.95, supporting more reliable schema harmonization for simulation-ready inputs. Full article