Geographic knowledge graphs (GeoKGs) represent semantic entities and relations as structured triplets, enabling intelligent retrieval and semantic reasoning in geospatial domains. However, existing knowledge graph embedding methods focus primarily on the instance view, overlooking ontological constraints and semantic hierarchies. This oversight creates a semantic segregation between concepts and instances, thereby limiting model performance in tasks such as relation prediction and generalisation for low-frequency entities. To address these challenges, this paper proposes a joint ontology–instance embedding framework within a unified semantic space, achieving synergistic modelling of ontological concepts and instance entities in GeoKGs. Specifically, we adopt CompoundE as the foundational model and introduce a collaborative embedding mechanism that enforces type constraints and semantic alignment through geometric transformations, including translation, rotation and scaling. Furthermore, a joint loss function is designed to simultaneously optimise the ontology hierarchy, instance structure and cross-view consistency. Experimental results demonstrate that our proposed method significantly outperforms multiple state-of-the-art baselines in ranking accuracy and generalisation capability on link prediction tasks, validating the effectiveness and necessity of integrating the ontology view into geographic knowledge representation learning.

The rapid expansion of social media platforms and the fragmentation of interest communities have created new theoretical demands for understanding content propagation. Users routinely combine community-specific semantic vernaculars in their content, yet combination-level propagation patterns remain invisible within single-vernacular frameworks. This study constructs a color vernacular co-occurrence network from ACG-related posts on Xiaohongshu (January 2024 to March 2025) and examines how cognitive hierarchy compatibility and network structural position jointly associate with combination propagation outcomes. Three findings emerge. First, semantic hierarchy compatibility positively associates with propagation synergy: within-level compatible combinations correspond to higher propagation gains and cross-level incompatible combinations correspond to negative synergy. Second, cognitive distance associates with propagation outcomes in a monotonically accelerating declining pattern, which does not support the inverted-U hypothesis, because the cognitive gains from semantic novelty cannot offset the decoding burden that accelerates as hierarchical distance increases. Third, semantic hierarchy and community structure show opposite directional associations: cross-level semantics correspond to a negative association while cross-community bridging corresponds to a positive association, and cognitively compatible cross-community combinations consistently outperform incompatible ones. Maintaining semantic hierarchy compatibility while pursuing structural bridging, and incorporating cognitive compatibility into platform recommendation mechanisms, are productive directions for future research.

Aim. Prior research concentrated on testifying to the presence of the semantic activation of semantic radicals as well as the separate role of a couple of modulating factors like Transparency of the phonograms, Position of semantic radicals, and Authenticity of the phonograms. However, the combined and hierarchical effects of these factors on non-character semantic accessibility remain largely unexplored. The present study, first proposed a hypothesis on the semantic accessibility hierarchy of non-character semantic radicals, and empirically investigated the hypothesis. Materials and Design. The experiment adopted a 2 (Prime type: body effector vs. Asterisk *) × 6 (Target type: transparent phonograms embedding non-character effector semantic radical on the left [TPA1] vs. opaque phonograms embedding non-character effector semantic radical on the left [TPA2] vs. pseudo-characters embedding non-character effector semantic radical on the left [TPA3] vs. pseudo-characters embedding non-character effector semantic radical on the right [TPA4] vs. TPAcontrol1 vs. TPAcontrol2) design. The prime type comprised three effector-based characters: “脚” (/jiao3/, foot), “嘴”(/zui3/, mouth), and “手” (/shou3/, hand), together with a control asterisk “*”. A primed lexical decision task was administered to 30 native Mandarin-speaking participants. Results. Data of 26 participants were retained for analysis, yielding a mean accuracy of 93.62%. Trials with errors or reaction times exceeding 3 SD were excluded prior to analysis. Significant priming effects were observed across most target conditions, with effect magnitudes partially confirming the proposed hierarchy: the highest semantic accessibility was observed for non-character semantic radicals left-positioned in pseudo-characters (TPA3), followed by those right-positioned in pseudo-characters (TPA4), then those left-positioned in transparent phonograms (TPA1), and the lowest for those left-positioned in opaque phonograms (TPA2). Conclusions. The semantic behavior of non-character semantic radicals was shaped by the host phonograms configured by such modulating factors as phonograms’ transparency, authenticity, and semantic radicals’ position.

In robotic fine-grained tasks (e.g., grasping and assembly), precise interaction requires a detailed understanding of object components. While Visual Language Models (VLMs) excel at object-level recognition, they struggle with part-level segmentation (e.g., knife handles), limiting performance in complex scenarios. VLMs face three key challenges: (1) Visual granularity mismatch-object-level features lack part-level details; (2) Semantic hierarchy gaps-parts and objects differ significantly in semantics; (3) Cross-modal bias-CLIP's text-image alignment favors global over local features. To address these, we propose a one-stage VLM-based part segmentation method. First, the Hierarchy-Aware Feature Selection mechanism analyzes Transformer features in different hierarchies to enhance spatial and semantic precision for part segmentation. Second, the Multi-Hierarchy Feature Adapter bridges object-to-part feature granularity via the hierarchical adaptation. Finally, the Hierarchical Multimodal Alignment Module harmonizes classification accuracy and mask integrity via hierarchical alignment of vision-language, mitigating the bias of CLIP's object-level priori knowledge. Experiments show the proposed method improves part segmentation performance for Zero-Shot, achieving 25.86% on Pascal-Part and 13.09% on ADE20K-Part (gains of +0.81% hIoU and +2.96% hIoU over baseline). This work advances robotic visual perception, with applications in intelligent manufacturing and intelligent service.

Achieving lightweight representations of building mesh models with accurate geometry and fine structural details is a key challenge in urban 3D modelling. Most existing mesh simplification methods focus on minimizing geometric error while neglecting the specific characteristics of building models in terms of geometric structure and semantic hierarchy, thus leading to structural degradation and semantic inconsistencies. To address this issue, this paper proposes a structure–semantic dual-constrained edge-collapse decimation method for simplifying dense building mesh models reconstructed from point clouds. Our core innovation lies in the joint enforcement of geometric structural constraints and building semantic constraints to effectively preserve both geometric structural features and component-level semantic structures of the models. By incorporating these two constraints, we adaptively assign higher collapse penalties to key structural edges and semantic boundaries, achieving lightweight building model simplification while maintaining fine-level structural details even under high compression ratios. Our method is extensively validated on several datasets of varying scales and complexities, including single-building models from Sketchfab, the large-scale urban datasets SUM and STPLS3D, and the ArCH cultural heritage dataset. Experimental results demonstrate that our method achieves superior or comparable performance compared to the existing methods across all the test datasets, consistently achieving lower or on-par geometric errors measured by RMSE and MAE. Furthermore, our simplified results can be semantically organized and stored under the CityGML paradigm, which provides a unified data support for sharing, semantic retrieval, downstream analysis, and other applications of lightweight building models.

This article analyzes the semantic structure of Internet comments and uses A. Nurmonov’s concept of “semantic content of a lexeme” as a theoretical basis. Sema is interpreted as a hierarchical phenomenon in a five-level classification — nominative, significant, structural, connotative and pragmatic layers. Therefore, Internet discourse is interpreted as a generalization combining lexical, cognitive and pragmatic meanings based on semantic hierarchy.

This study explores the syntactic linguopoetics of declarative sentences and examines their structural, semantic, and expressive potential in literary texts. Declarative sentences, traditionally regarded as units that convey information, are analyzed here as multifunctional syntactic constructions capable of expressing emotional, aesthetic, and stylistic nuances. The research focuses on the correlation between syntactic form and syntactic meaning, the role of modality, and the invariant declarative seme that unites various semantic variants such as message, wish, confidence, advice, desire, agitation, existence, naming, and deixis. Special attention is paid to intonational features, standard and nonstandard word order, and the expressive possibilities created through inversion, repetition, paired synonyms, syntactic gradation, ellipsis, antithesis, and other stylistic devices. The analysis demonstrates that changes in word order and grammatical connections influence semantic hierarchy and foreground specific sentence elements, thereby intensifying meaning and enhancing expressiveness. The findings confirm that declarative sentences function not only as carriers of factual information but also as key syntactic means of linguopoetic organization in artistic discourse.

Salient object detection in RGB-D imagery remains challenging due to inconsistent depth quality and suboptimal cross-modal fusion strategies. This paper presents a novel dual-stream architecture that integrates contextual feature refinement with adaptive attention mechanisms for robust RGB-D saliency detection. We extract two features from the ResNet-50 backbone for both the RGB and depth streams, capturing low-level spatial details and high-level semantic representations. We introduce a Contextual Feature Refinement Module (CFRM) that captures multi-scale dependencies through parallel dilated convolutions, enabling hierarchical context aggregation without substantial computational overhead. To enhance discriminative feature learning, we employ channel attention for inter-channel recalibration and a modified spatial attention mechanism utilizing quadruple feature statistics for precise localization. Recognizing that existing depth maps in benchmark datasets are outdated and degraded in quality, we introduce refined depth maps generated with Depth AnythingV2, which significantly improve cross-modal alignment and detection performance. The progressive fusion strategy integrates complementary RGB and depth information across semantic hierarchies, while the saliency prediction block generates high-resolution predictions via gradual spatial expansion. Extensive experiments across six benchmark datasets validate our approach, achieving competitive performance with recent state-of-the-art methods.

Clinical intent for laboratory testing ("indication") is rarely recorded in structured form, limiting contextual interpretation, auditability, and utilization stewardship. We developed a comprehensive, and clinically applicable framework that standardizes laboratory test indications and links them to indication-dependent utilization and interpretation. A structured literature review on utilization, appropriateness, and request rationale informed an iterative, consensus-based process with a multidisciplinary expert panel to develop and operationalize an indication taxonomy and attribute schema. Structural coherence was assessed by comparing semantic distance hierarchies derived from indication labels alone with an enriched multi-layer ("layered prototype") representation incorporating these attributes. Use cases were applied to assess feasibility of indication-to-interpretation mapping. We defined 19 distinct indication types, grouped into five clusters across the clinical course: Initial Detection and Diagnostic Clarification, Disease Characterization and Prognosis, Therapy Guidance and Safety, Longitudinal Management and Reassessment, and Analytical and External Requirements. Each is specified with structured attributes and examples to support implementation. Semantic distance analyses supported a coherent hierarchy. Layered prototypes yielded more informative organization than labels alone, enabling context-dependent consolidation and guided deployment. By providing explicit indication-to-interpretation mapping/logic, the framework closes a key gap in the total testing process between order entry and post-analytical interpretation. It supports context-specific decision limits, reporting logic, and stewardship analytics, and is amenable to formalization as a machine-readable ontology for interoperable implementation.

In the realm of real-world knowledge graphs, the dynamism of facts is a prevailing characteristic. To illustrate, a popular restaurant was awarded a Michelin star in 2004 and retained this prestigious recognition in 2008, but lost it in 2012 due to changes in management and menu quality. This sequence highlights how neglecting temporal context can lead to misconceptions about factual accuracy. Furthermore, the relations intertwining distinct entities or the same entity across different chronological markers exhibit complexity and hierarchy. Regrettably, existing methods for temporal knowledge graph relation prediction fall short in following challenges: they lack a nuanced, hierarchical comprehension of knowledge structure and fail to adeptly integrate temporal dynamics with static attributes. Addressing these issues, this study introduces Hyperbolic-based Temporal Knowledge Graph Relation Prediction (HTKGP) approach to harness the power of hyperbolic geometry. Our innovation is an attention-guided, learnable curvature mechanism designed to preserve and enrich the intricate semantic hierarchy inherent in data. Besides, we propose a longitudinal information entity embedding strategy due to the plentiful temporal information. This not only captures the enduring impact of past events on present states but also achieves efficiency through parameter reduction. Empirical validation across multiple datasets shows HTKGP efficiently navigates the rich semantic landscape within hyperbolic spaces and yields superior predictive performance. Our implementations are publicly available at: https://github.com/jianruichen/HTKGP .

Tunnel point cloud semantic segmentation is a critical step in achieving refined perception and intelligent management of tunnel structures. Addressing common challenges including indistinct boundaries and fine-grained category discrimination, this paper proposes MFPNet, a multi-scale feature perception network specifically designed for tunnel scenarios. This approach employs kernel convolution to effectively model local point cloud geometries within continuous spaces. Building upon this foundation, an error-feedback-based local-global feature fusion mechanism is designed. Through bidirectional information exchange, higher-level semantic information compensates for and constrains lower-level geometric features, thereby mitigating information fragmentation across semantic hierarchies. Furthermore, an adaptive feature re-calibration and cross-scale contextual correlation mechanism is introduced to dynamically modulate multi-scale feature responses. This explicitly models contextual dependencies across scales, enabling collaborative aggregation and discriminative enhancement of multi-scale semantic information. Experimental results on tunnel point cloud datasets demonstrate that the proposed MFPNet has achieved significant improvements in both overall segmentation accuracy and category balance, with mIoU reaching 87.5%, which is 5.1% to 33.0% higher than mainstream methods such as PointNet++ and RandLA-Net, and the overall classification accuracy reaching 96.3%. These results validate the method’s efficacy in achieving high-precision three-dimensional semantic understanding within complex tunnel environments, providing robust technical support for tunnel digital twin and intelligent detection applications.

This article analyzes the specific aspects of the conversion phenomenon in the Uzbek language. The study covers in detail semantic hierarchy of converted words, their internal structural model, and the process of integrative meaning formation. During the analysis, the article focuses on the semantic relationship of the conversions, that is, the first is determinative, and the second form is the main (semantic center) function. As a result, their ability to create a new lexical meaning as a semantic unit is emphasized as the main conclusion.

Gait recognition offers non-contact, long-distance identification but struggles with robustness against covariates like clothing variations, carrying conditions, and viewpoint changes. Existing methods predominantly rely on single modalities (e.g., silhouettes or skeletons) or employ shallow multimodal fusion, such as simple concatenation, which treats modalities as independent and static, failing to exploit their complementary strengths, shape cues from silhouettes and structural kinematics from skele-tons. To address these limitations, we introduce the Synergistic co-evolving representations (See) principle, enabling modalities to iteratively interact, guide, and refine each other across semantic hierarchies, fostering a unified, robust identity representation resilient to complex environments. This is realized through SeeGait, a novel multimodal framework featuring hierarchical multi-stage fusion. At its core, the Bidirectional Hierarchical Cross-Attention Synergy Module (BiHCASM) employs adaptive cross-modal attention to dynamically align and reweight features bidirectionally, allowing structural insights to enhance appearance focus and vice versa. Complementing this, the Hierarchical Spatiotemporal Transformer Encoder (HSTE) captures long-range skeleton dynamics, overcoming GCN limitations, while the Hierarchical Convolutional Silhouette Encoder (HCSE) extracts multi-scale silhouette pyramids for rich shape priors. Finally, a Holistic Feature Aggregation (HFA) strategy consolidates features from all stages for deep supervision, ensuring comprehensive optimization. By promoting mutual refinement, SeeGait mitigates covariate disruptions through enhanced complementarity, yielding superior discriminability. Extensive experiments show state-of-the-art performance, with 97.1% average Rank-1 accuracy on CASIA-B, and top results on CCPG and SUSTech1K.

Radiology report generation (RRG) has emerged as a promising approach to alleviate radiologists' workload and reduce human errors by automatically generating diagnostic reports from medical images. A key challenge in RRG is achieving fine-grained alignment between complex visual features and the hierarchical structure of long-form radiology reports. Although recent methods have improved image-text representation learning, they often treat reports as flat sequences, overlooking their structured sections and semantic hierarchies. This simplification hinders precise cross-modal alignment and weakens RRG accuracy. To address this challenge, we propose RIHA (Report-Image Hierarchical Alignment Transformer), a novel end-to-end framework that performs multi-level alignment between radiological images and their corresponding reports across paragraph, sentence, and word levels. This hierarchical alignment enables more precise cross-modal mapping, essential for capturing the nuanced semantics embedded in clinical narratives. Specifically, RIHA introduces a Visual Feature Pyramid (VFP) to extract multi-scale visual features and a Text Feature Pyramid (TFP) to represent multi-granularity textual structures. These components are integrated through a Cross-modal Hierarchical Alignment (CHA) module, leveraging optimal transport to effectively align visual and textual features across various levels. Furthermore, we incorporate Relative Positional Encoding (RPE) into the decoder to model spatial and semantic relationships among tokens, enhancing the token-level alignment between visual features and generated text. Extensive experiments on two benchmark chest X-ray datasets, IU-Xray and MIMIC-CXR, demonstrate that RIHA outperforms existing state-of-the-art models in both natural language generation and clinical efficacy metrics.

Unsupervised Domain Adaptation (UDA) has emerged as a pivotal technique for enhancing machine learning models' performance in unlabeled target domain with domain shifts. This technique is fundamentally achieved by aligning the domain distributions of source and target domains within a latent feature space, thereby enhancing model robustness across heterogeneous data distributions. However, the inherent discrepancy between source and target domain distributions poses significant challenges in identifying the optimal latent space. Furthermore, projecting both domains into suboptimal latent spaces may induce substantial semantic information loss, particularly compromising discriminative feature representations critical for final tasks. In this article, our systematic analysis reveals that natural language representations inherently possess stronger semantic abstraction capabilities than visual features in natural images. As a result, natural language tends to have smaller domain shifts. Motivated by this discovery, we proposed a novel model that systematically transforms visual patterns into structured linguistic representations. This cross-modal translation mechanism leverages the invariant semantic properties of natural language to mitigate domain shifts while preserving task-critical semantic hierarchies. Our model leverages the inherent abstraction capacity of linguistic structures to enhance cross-domain generalization, effectively bridging the visual-semantic gap in unsupervised adaptation scenarios. Our model comprises three core components: 1) text classification branch translating images to text for prediction; 2) image adaptation branch supplementing visual details; and 3) ensemble mechanism reconciling text abstraction with visual granularity through mismatch detection. Extensive experiments on three benchmark datasets validate the effectiveness of our model, achieving state-of-the-art performance.

Existing deep learning-based cross-view geolocalization methods primarily focus on improving the accuracy of cross-domain image matching. Less attention is paid to ensuring that models can comprehensively capture contextual information around the target and minimize the cost of localization errors. To support quantitative research into this Distance-Aware Cross-View Geo-Localization (DACVGL) problem, we construct Distance-Aware Campus (DA-Campus), the first benchmark that pairs multi-view imagery with precise distance annotations across three spatial resolutions. Based on DA-Campus, we formulate DACVGL as a hierarchical retrieval problem across different domains. In this setting, we further identify that due to the inherent complexity of spatial relationships among buildings, conventional metric learning lacks a unified semantic hierarchy to guide the organization of the latent feature space. To tackle this challenge, we propose Dynamic Contrastive Learning (DyCL), a novel framework that progressively aligns feature representations according to spatial margins. Extensive experiments demonstrate that DyCL can serve as a strong baseline for the DACVGL task, yielding substantial improvements in both hierarchical retrieval performance and overall geo-localization accuracy. Our code and benchmark are publicly available at https://github.com/anocodetest1/DyCL.

Automatic polyp segmentation is a key task in colonoscopic image analysis. It provides the basis for polyp size estimation, boundary localization, and morphological analysis to support clinical decision-making. However, variations in polyp shape and size and low image contrast often lead to unstable boundary delineation and spatial localization, resulting in structural measurement bias. To address this issue, we propose DSCNet, a polyp segmentation network designed to improve structural measurement stability by jointly modeling hierarchical semantic features and dimensional characteristics. For low- and mid-level features, a Channel Synergistic Local Enhancement (CSLE) module is introduced, which combines local convolutional modeling with channel-wise visual cues to enhance boundary discrimination and multi-scale morphological consistency. For high-level features, a Multi-scale Spatial Perception Regulation (MSPR) module is designed to strengthen the perception of low-contrast regions through directional and multi-scale spatial modeling, thereby improving the reliability of polyp size estimation and spatial localization. In addition, a Hierarchical Attention-guided Fusion (HAF) module is employed to alleviate semantic discrepancies across feature levels and further refine boundary structures. Experimental results on five public polyp datasets demonstrate that DSCNet consistently outperforms existing methods. In particular, on the challenging ETIS dataset dominated by small polyps, DSCNet achieves approximately a 1.8% relative improvement in Dice and an 8.0% relative reduction in HD95 compared with the best baseline, indicating superior spatial localization stability and boundary accuracy.

In this paper, I define and develop the concept of “(textual) anaphora”, i.e. text relations between an anaphor and an antecedent, and I focus particularly on the so-called direct anaphors, viz. the anaphors that designate the same individual entity as the antecedent. My primary attention is on the linguistic material of the anaphors, which fundamentally depends on the antecedent’s pragmatic prominence, i.e. its degree of presence and saliency in the hearer’s mental-cognitive representation, at the moment it is anaphorised. This prominence is based partly on the antecedent’s topicality, which in turn depends on its place in four different semantic, syntactic and referential hierarchies, partly on the text and narrative structure, e.g. the textual distance between antecedent and anaphor, the presence of other possible antecedents, and possible topic changes. The topicality and the text/narrative structure can also be defined as “vertical” and “horizontal” criteria respectively, and as empirical evidence I use a bilingual (Italian – Danish) text corpus as well as other text sources. All examples are translated into English.

This study addresses the problem of generalized category discovery (GCD), an advanced and challenging semi-supervised learning scenario that deals with unlabeled data from both known and novel categories. Although recent research has effectively engaged with this issue, these studies typically map features into Euclidean space, which fails to maintain the latent semantic hierarchy of the training samples effectively. This limitation restricts the exploration of more detailed and rich information and degrades the performance in discovering new categories. The emerging field of hyperbolic representation learning suggests that hyperbolic geometry could be advantageous for extracting semantic information to tackle this problem. Motivated by this, we proposed hyperbolic hierarchical representation learning for GCD (HypGCD). Specifically, HypGCD enhances representations in hyperbolic space, building upon the Euclidean space representation from two perspectives: instance-class level and instance-instance level. At the instance-class level, HypGCD endeavors to construct well-defined clusters, with each sample forming a robust hierarchical cluster structure. Concurrently, at the instance-instance level, HypGCD anticipates that a subset of samples will display a tree-like structure in local space, which aligns more closely with real-world scenarios. Finally, HypGCD optimizes the Euclidean and hyperbolic space collectively to obtain refined features. Additionally, we show that HypGCD is exceptionally effective, achieving state-of-the-art (SOTA) results on several datasets. The code is available at https://github.com/DuannYu/HypGCD.

ABSTRACT While the rich semantic information in trajectories enhances data mining potential, it simultaneously complicates similarity measures. Existing multidimensional similarity methods face two challenges: (1) high computational complexity of O(n × m × k), which limits large-scale applicability, and (2) isolated handling of spatial–temporal-semantic dimensions with inadequate semantic hierarchy modeling. This paper proposes FasMultiSIM, a fast and synchronized multidimensional similarity measure that integrates spatial, temporal, and semantic dimensions. First, a spatiotemporal grid model (rSTGM) is constructed by extending rHEALPix DGGS with temporal dimensions, enabling FasMultiSIM to achieve fast multidimensional similarity computations with time complexity of O((n + m)×k). In addition, FasMultiSIM quantifies hierarchical semantic relationships while preserving the flexibility of semantic tags and enables synchronized multidimensional similarity measures to capture interesting and valuable segments in similar trajectories. Finally, we validated the accuracy and efficiency of FasMultiSIM using both real-world floating cars and social media trajectory datasets. The experimental results demonstrate that FasMultiSIM achieves an approximately one order of magnitude improvement in computational efficiency compared to the state-of-the-art methods, including LCSS, EDR, MSM, and MUITAS. The proposed method is expected to support applications that demand high timeliness and focus on synchronized multidimensional similarity, such as epidemic tracking and personalized intelligent recommendation services.