ABSTRACT Joint clustering of hyperspectral and Light Detection and Ranging (LiDAR) data is challenging due to their heterogeneity and differing spatial-spectral characteristics. To address this, we propose an adaptive multi-view graph convolutional network (MVGCN) that integrates visual Bidirectional Encoder Representations from Transformers (VisualBERT), referred to as MVGCN-VisualBERT, to extract high-level semantic features from both modalities. These features form a superpixel-level graph that preserves spatial structure while reducing redundancy. A multi-view graph convolutional network then propagates and aggregates information to enhance cluster cohesion. Evaluated on the MUUFL and UH2013 datasets, MVGCN-VisualBERT outperforms state-of-the-art methods, achieving improvements of 2.8% in overall accuracy, 2.6% in the Kappa coefficient, 2.9% in normalized mutual information and 5.8% in the adjusted Rand index on MUUFL. These results highlight the potential of the proposed approach for improving unsupervised multimodal land-cover analysis in remote sensing applications.

Multimodal metaphors in animated depression educational videos on Chinese social media: A critical multimodal discourse analysis.

Decoding evolution of chemical and sensory profiles in Vitis quinquangularis fortified wines during bottle aging via multimodal analysis

MulMoSenT: Multimodal sentiment analysis for a low-resource language using textual-visual cross-attention and fusion

DGFN: Disentanglement-guided dynamic fusion network for multimodal sentiment analysis

Automatic emotion recognition plays a critical role in areas such as mental-health monitoring, human–robot interaction, and personalised learning systems, yet current multimodal approaches often struggle with high intra-class variability and the limited discriminative power of raw audio–visual features. Existing methods typically rely on direct classification of audio or facial data, which does not explicitly enforce a structured joint embedding in which emotional categories become separable. This paper addresses this limitation by proposing a supervised contrastive feature-mapping algorithm that transforms temporal audio and video features into a representation that minimises intra-class distances while maximising inter-class distances. In contrast to prior work, which usually focuses on handcrafted feature engineering or end-to-end classifiers, our approach explicitly learns a discriminative metric space that enhances the geometry of the feature distribution. The method is evaluated on the RAVDESS and CREMA-D benchmark datasets. Experimental results show that the proposed mapping yields consistent accuracy improvements over strong machine-learning baselines, with gains of up to approximately 6%, achieving 96.07% accuracy on RAVDESS and competitive performance on CREMA-D, while outperforming or matching recent state-of-the-art multimodal emotion-recognition pipelines. Statistical tests (Kruskal–Wallis and paired t-tests) confirm that the learned representation significantly increases class separability ( ). While the method assumes the availability of paired audio–visual inputs without requiring explicit temporal alignment, the learned feature space is compact, discriminative, and well suited to downstream tasks such as affect-aware dialogue systems, rehabilitation monitoring, and adaptive educational interfaces. These results demonstrate that contrastive feature mapping provides a robust and generalisable framework for multimodal emotion analysis. • We propose a supervised contrastive model that maps temporal audio-visual features into a joint representation, minimizing intra-class variations to enhance the separability of emotional classes in the transformed feature space. • We provide an in-depth analysis of feature distributions before and after transformation, demonstrating improved class discrimination. • Our algorithm achieves superior performance compared to existing approaches, and we publicly release the software for further research and development.

Grading-inspired complementary enhancing for multimodal sentiment analysis

Decoding the dynamic formation of bolus: A multimodal fusion and simulation framework for texture perception-regulated oral processing.

Fusion decomposition and backbone gathering based multimodal sentiment analysis under uncertain missing modalities

Multimodal sentiment analysis based on noise injection and adaptive multi-view dynamic fusion

Neuroimaging, particularly magnetic resonance imaging (MRI), has become a cornerstone in elucidating the neural underpinnings of Major Depressive Disorder (MDD). As the volume of related literature grows rapidly, a systematic, quantitative overview of the field's development, intellectual structure, and emerging trends is essential. This study conducted a scientometric and visual analysis of MRI research on MDD over the past 28 years (1997-2025) using data extracted from the Web of Science Core Collection. A total of 3269 publications were analyzed with CiteSpace and VOSviewer to map collaboration networks, thematic evolution, and research fronts. The results show a steady increase in annual output, particularly after 2006, driven by advances in 3T MRI and the pursuit of objective biomarkers. The United States and China are the most productive countries, while Harvard University and the University of California System lead institutional contributions. Co-citation and keyword analyses reveal a paradigm shift from localized structural/functional deficits toward network-based perspectives, alongside methodological evolution toward multimodal integration, dynamic analyses, and machine learning. Current research fronts focus on suicidal ideation, treatment prediction, and neuromodulation. This study provides a macroscopic overview of the field's trajectory, highlighting key contributors, thematic transitions, and future challenges in translating neuroimaging findings into clinical practice.

AsyCMST: Asymmetric cross-modal spatio-temporal learning for multimodal ultrasound nodule recognition.

LLM-based agent for urban sewer pipeline inspection integrating image enhancement and multimodal defect analysis

Integrating actigraphy with demographic data enhances cognitive performance prediction: a multimodal UK biobank analysis using machine learning.

DCAF: Dynamic affective consistency-aware fusion with disentangled modality representations for multimodal sentiment analysis

From diagnostic labels to radiology reports: a unified multi-modal framework for lesion detection and segmentation

Changes in kynurenine pathway metabolites and brain structure in response to behavioral activation therapy.

Repetitive injury is hypothesized to lead to progressive tissue fibrosis and end-stage organ failure. Whether tissue-resident mesenchymal cell populations retain epigenetic memory of prior injuries that contribute to this pathological process is unknown. Here we used a genetic lineage labeling approach to mark the lung mesenchyme prior to injury, then performed multimodal analyses on isolated lung mesenchyme during the initiation, progression, and resolution of the fibrotic response. Our results demonstrate the remarkable epigenetic and transcriptional plasticity of the lung mesenchyme during fibrotic activation and de-activation. Despite this plasticity, we also find that the lung mesenchyme exhibits an enhanced fibrotic program upon reinjury. We identify RUNX1 as a critical driver of both fibrotic activation and fibrotic memory. Comparison of fresh isolated and cultured lung mesenchyme demonstrates that RUNX1 is spontaneously activated in standard culture conditions, previously masking these roles of RUNX1. Targeted knockdown of RUNX1 dampens fibrotic mesenchymal cell activation immediately after cell isolation, but with reduced efficacy after only days of culture, confirming its functional importance to both early activation and long-term memory. Collectively, our findings implicate RUNX1 in the initiation and memory of fibrotic mesenchymal cell activation that together prime enhanced mesenchymal cell responses upon repeated injury.

This article examines how social reality is constructed in short-form documentary films by comparing the materials of NYT Op-Docs and Azattyq. Contemporary digital platforms increasingly transform a documentary film into an epistemological media form that selects, organizes, and interprets public events. Existing studies on documentary film, narrative journalism, and visual framing have mainly focused on either Western media contexts, thematic representation, or aesthetic strategies. However, there is still limited comparative research on how institutionally and geographically different media platforms construct reality through distinct documentary logics. This study addresses that gap by analyzing short-form documentaries as platform-specific systems of knowledge production. The purpose of the article is to identify the mechanisms through which short documentary films organize social reality and to define the dominant epistemological models operating within two different media environments. The theoretical framework combines framing theory, narrative journalism, and multimodal discourse analysis. The empirical corpus consists of ten short documentary films selected through purposive sampling: five films from NYT Op-Docs and five films from Azattyq. The films were divided into 45 narrative segments and analyzed through qualitative content analysis and multimodal analysis. The coding scheme included five categories: narrative structure, type of evidence, visual mode, institutional visibility, and affective organization. The findings demonstrate two contrasting models of documentary reality construction. NYT Op-Docs primarily operates through a narrative-affective model, in which social reality is organized around personal experience, character-centered storytelling, aesthetic visual composition, and emotional reflection. Azattyq, by contrast, follows an evidentiary-testimonial model, where reality is constructed through witness accounts, factual evidence, raw visual material, and explicit institutional responsibility. The article contributes to framing theory by extending it beyond textual and news-based analysis and by showing how documentary meaning is produced through the interaction of narrative, visual, evidentiary, institutional, and affective elements. The study also demonstrates that short-form documentary films should be analyzed not only as journalistic products, but as epistemological mechanisms that shape public understanding of social reality.

Hard carbon (HC) anode in sodium-ion batteries suffer from low initial Coulombic efficiency and irreversible capacity loss, limiting practical energy density and cycle life of SIBs. While direct-contact presodiation of HC has been proposed to increase the initial Coulombic efficiency of SIBs, but its low utilization efficiency can cause residual Na on the HC surface, resulting in rapid degradation and even safety concerns. Herein, we proposed a soft-contact presodiation (SCP) method, which can remove and recycle Na source and therefore greatly improve the utilization of the Na source and safety of SIBs. The SCP-treated HC anode achieves a ≈30.0% increase in ICE when paired with a NaNi1/3Fe1/3Mn1/3O2 cathode, while maintaining minimal temperature rise (ΔT≈1.3°C) during treatment. The resulting SCP-HC exhibits exceptional thermal stability with negligible exothermic activity at 125.0°C and remains chemically stable for over 3.0 days. Through multimodal analysis, we reveal an adsorption-dominated compensation mechanism where replenished Na participates in solid electrolyte interphase formation while simultaneously occupying adsorption sites as metallic clusters. The pouch cell incorporating SCP-HC anode delivers 90.6% ICE and retains 80.0% capacity after 150 cycles. This work establishes a safe, efficient, and economically viable presodiation platform that paves the way for practical high-energy sodium-ion batteries.