The widespread dissemination and misleading impact of fake news on the web have become a significant concern for the public and the government. Discovering fake news is crucial for ensuring that users receive authentic information and maintaining social harmony. However, most existing entity-based fake news detection methods have two issues: i) methods for acquiring additional information through entities lack flexibility and real-time capabilities. ii) approaches using entities to capture news semantics have not adequately revealed the interactions between words in the text. To address these issues, we propose a <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><b>M</b></u>ulti-<underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><b>g</b></u>raph <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><b>S</b></u>emantic-aware <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><b>A</b></u>daptive Graph Convolutional <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><b>N</b></u>etwork (<bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">MgSAN</b>), which comprehensively captures the semantic information of news texts by constructing multiple semantic graphs and learns the features from these graph structures using an adaptive graph convolutional network (SwiGCN). Specifically, we design a global semantic interaction graph to capture the complex interactions between words, generating a comprehensive textual semantic representation. We also employ an entity-noun relationship graph to mine deep semantic associations, enhancing the model's understanding of fine-grained textual deep meanings. Additionally, we develop an adaptive graph convolutional network to effectively extract and aggregate feature information from different graph structures. Finally, we introduce a fusion module to integrate both global and local fine-grained semantic information, forming a rich composite semantic representation, thereby improving the effectiveness of fake news detection. Extensive experimental results on three public benchmark datasets verify the effectiveness and superior performance of MgSAN, outperforming state-of-the-art detection models.

Wildfire detection and localization in aerial imagery is critical for rapid response and damage mitigation. Autonomous aerial vehicles (AAVs) enable large area monitoring but face real-time processing challenges due to limited onboard computational and sensor resources. This work introduces a cross-modal knowledge distillation framework for edge-deployed AAVs. A teacher network trained only on thermal images transfers semantic and spatial representations to an optical image based student network when trained in an offline fashion using thermal and optical image pairs. During deployment, the student uses only optical images, thus reducing reliance on multi-sensor payloads while maintaining high detection accuracy. The student model incorporates dual classification heads: an image-level head for fire-free vs. fire-impacted scenes, and a patch-level head for flame vs. no-flame discrimination. This patch-level strategy provides effective fire localization while avoiding the computational overhead of segmentation, making it practical for resource-constrained deployment. Evaluated on aerial wildfire dataset, the framework achieves 90.97% patch-level accuracy, with false alarm and missed detection rates of 8.82% and 14.78%, respectively. The lightweight student model requires only 2.99 GFLOPS with inference time of 0.004s and generates patch-level probability heatmaps for fire region localization. Unlike conventional unimodal systems, this approach leverages thermal-to-optical knowledge transfer to deliver high accuracy, low latency, and precise localization under edge-computing constraints. The code and dataset will be released at https://github.com/medh132/cmkd .

Neuroscience Prior knowledge guided EEG representation disentanglement for auditory attention decoding.

Time-frequency contrastive learning with context modeling for time series anomaly prediction.

EOAC-LLM: An LLM-driven event ontology automatic construction system

<b>Context: </b>Log anomaly detection is critical for maintaining the security, stability, and operational efficiency of modern software systems, especially as they generate vast and diverse log data. However, existing deep learning models struggle with the challenges of heterogeneous log formats across systems and the scarcity of labeled anomaly logs, limiting their real-world deployment and generalization capabilities. <br/><b>Objective: </b>To address these challenges, we propose LogMeta, a novel semi-supervised framework designed for adaptive and efficient log anomaly detection in diverse and low-resource environments. <br/><b>Method: </b>LogMeta integrates Model-Agnostic Meta-Learning (MAML) with a hybrid language model to address key challenges. MAML enables LogMeta to rapidly adapt to unseen log systems using few-shot samples, while the hybrid model combines RoBERTa for extracting semantic representations with Bi-LSTM and attention mechanisms to capture sequential dependencies and critical features within log sequences. This design reduces reliance on large-scale labeled datasets and enhances adaptability in heterogeneous environments. <br/><b>Results: </b>Experimental evaluations on multiple benchmark datasets demonstrate that LogMeta consistently outperforms state-of-the-art supervised and unsupervised methods, achieving up to a 28.3% improvement in F1-scores under low-resource scenarios compared to other models. Furthermore, LogMeta exhibits exceptional domain transfer capabilities, maintaining robust performance across diverse log datasets with minimal fine-tuning. In terms of efficiency, LogMeta achieves competitive training and inference times, making it suitable for real-time anomaly detection in large-scale systems. <br/><b>Conclusion: </b>LogMeta provides a scalable and practical solution for real-world log anomaly detection, overcoming challenges related to data heterogeneity and label scarcity. Its strong generalization capabilities, minimal supervision requirements, and adaptability to new log systems make it a promising tool for enhancing software system reliability and security. <br/>© 2026 The Author(s).

Diverse semantic representation learning based on vision-language models for zero-shot indoor scene recognition

Managing large volumes of documents remains a challenging task for individuals and organizations, especially when data exists in unstructured formats such as scanned files, PDFs, and images. Traditional document management approaches rely heavily on manual effort for sorting, searching, and extracting information, which leads to inefficiencies and increased processing time. To address these challenges, this paper introduces Neural Docs, an AI-enabled document management system that focuses on automating document understanding and interaction. The system is designed to transform static documents into interactive and searchable knowledge sources. It utilizes Optical Character Recognition (OCR) to convert visual document content into machine-readable text. This extracted information is further processed using Natural Language Processing (NLP) techniques to identify key entities, generate metadata, and organize documents intelligently. In addition, a Retrieval-Augmented Generation (RAG) mechanism is incorporated to enable users to query documents through a conversational interface, providing responses based on relevant contextual information rather than simple keyword matching. The overall architecture is divided into multiple layers, including a user interface for interaction, a backend system for processing and coordination, and a dedicated AI module for advanced analysis. A vector-based storage mechanism is used to maintain semantic representations of documents, allowing efficient similarity-based retrieval. The system is implemented using modern technologies such as Node.js, FastAPI, and containerized deployment for flexibility and scalability. The developed solution demonstrates improved accessibility and reduced manual workload in document handling tasks. It allows users to quickly retrieve meaningful insights from stored documents and interact with them in a more intuitive way. The approach presented in this work highlights the potential of combining multiple AI techniques to create smarter and more efficient document management systems suitable for real-world applications

In bandwidth-limited and time-varying vehicle–road–cloud cooperative autonomous driving scenarios, real-time transmission and joint inference of high-dimensional multimodal perception data are simultaneously constrained by latency, reliability, and energy consumption. To address these challenges, this article proposes a task-oriented multimodal fusion framework named Multi-Agent Dynamic Diffusion Semantic Communication Network (MA-DDSCNet). On the vehicle side, we design a Task-Guided Multi-Modal Semantic Encoder (TG-MMSE) that performs spatio-temporal alignment, complementary memory gating, and differentiable discrete quantization to compress heterogeneous perception streams from cameras, Light Detection and Ranging (LiDAR), and vehicular state into task-weighted discrete token sequences. A hierarchical distillation scheme is further employed to maintain a unified semantic coordinate system across vehicles, Road Side Units (RSUs), and the cloud. On the communication side, a hierarchical controllable diffusion mechanism adaptively adjusts diffusion noise and time steps according to the importance of object detection, trajectory prediction, and motion planning tasks, as well as link-specific bandwidth budgets. A multi-agent deep scheduler enables collaborative utilization of communication resources among the cloud, RSUs, and vehicles, while an iterative joint semantic decoding and consistency calibration algorithm feeds residuals back into a global memory matrix to suppress semantic drift and yield isomorphic semantic representations at all three layers. Furthermore, we construct a learnable uncertainty-driven multi-objective loss function, combined with a gradient projection strategy, to achieve end-to-end joint optimization of detection, prediction, and planning within a single training loop. Simulation results demonstrate that, compared with baseline methods, MA-DDSCNet achieves average gains of 9.6–18.4% in mean Average Precision (mAP), Average Displacement Error (ADE), Final Displacement Error (FDE), Effective Bit Rate (EBR), and planning safety rate, while reducing the 95th-percentile end-to-end latency to 63 ms, indicating that the proposed framework can significantly enhance the overall performance of semantic perception tasks in complex vehicular networks.

ABSTRACT The active expansion of digital media, as well as social networking services, has raised the dispersion of false information and fake news to serious issues, which put pressure on society, its opinion, and democracy itself. Manual verification is not easy and efficient since fake news tend to spread more rapidly than real news because it is sensational. The current research provides NEWSGUARD, a smart system that is aimed at the automated fake news detection relying on a hybrid method of using machine learning and deep learning techniques. The suggested system proposes the use of Natural Language Processing (NLP) in processing of text and extracting features. Term Frequency-Inverse Document Frequency (TF-IDF) is used to create statistical features whereas semantic representations are retrieved with word embeddings. Various classification models are deployed, such as Naive Bayes, Logistic Regression, Support Vector Machine (SVM), Convolutional Neural Networks (CNN) and Long short-term memory (LSTM). The hybrid model is created that combines the benefits of both machine learning and deep learning models and allows achieving better performance in classification. The experimental findings show that the hybrid model has a higher performance compared to the individual models where it has a high accuracy with a high precision, recall and F1 score. This research result demonstrates the usefulness of statistical and semantic features as a combination to detect fake news. The suggested NEWSGUARD solution offers the scalable, correct, and effective tool to fight with the misinformation, and it will be able to be further improved by using advanced models and real-time implementation.

This study develops a multi-loop knowledge management framework for capturing and refining knowledge from unstructured text. A system-oriented design is applied to 80 classical Chinese texts and interaction data from 200 users. The framework integrates automated extraction, semantic representation, and human-in-the-loop feedback, evaluated via baseline comparison, performance metrics, and robustness tests. Multi-source feedback improves topic coherence, interpretability, reduces latency, and stabilizes knowledge base construction. The study extends knowledge management constructs, including absorptive capacity, dynamic capabilities, and adaptive knowledge governance, through a multi-loop feedback mechanism. The framework enhances knowledge traceability and supports decision-making in knowledge-intensive contexts. This study proposes a configurable knowledge management system architecture integrating iterative feedback loops for managing complex knowledge assets.

The global e-commerce market is projected to surpass USD 7 trillion by 2030, with more than 80% of consumers relying heavily on product reviews to inform their purchasing decisions. Despite this reliance, manual sentiment analysis of such reviews is inherently limited in scalability and consistency, often leading to delayed insights and subjective or inaccurate interpretations. To overcome these limitations, this study introduces a comprehensive Natural Language Processing (NLP) framework designed for automated sentiment classification using a labeled product review dataset. The proposed approach begins with extensive preprocessing and Exploratory Data Analysis (EDA) to clean, standardize, and analyze the underlying data distribution. For advanced semantic representation, Sentence Bidirectional Encoder Representations from Transformers (SBERT) is utilized to generate context-rich embeddings, enabling the capture of nuanced linguistic patterns beyond the capabilities of traditional feature extraction methods. To address the issue of class imbalance across sentiment categories, the Synthetic Minority Over-sampling Technique (SMOTE) is applied to create synthetic samples, thereby ensuring balanced and effective model training. In contrast to conventional machine learning models such as Random Forest Classifier (RFC), Light Gradient Boosting Machine (LGBM), and Extreme Gradient Boosting (XGBoost), the proposed framework integrates a Deep Neural Network (DNN)-based feature selection mechanism with a Boosted Rules Classifier (BRC). This hybrid architecture not only enhances predictive accuracy but also improves model interpretability. The system categorizes customer sentiments into three distinct classes: Negative, Neutral, and Positive. Experimental findings demonstrate that the proposed approach achieves higher classification accuracy with reduced bias, validating its effectiveness. The framework offers a scalable and reliable solution for sentiment analysis, empowering businesses to make data-driven decisions in areas such as product development, marketing strategy optimization, and customer relationship management (CRM), ultimately enabling more responsive and customercentric operations.

Quantitative analysis of retinal vascular morphology is vital for clinical decision-making and the investigation of systemic diseases. Central to this process is the accurate segmentation of retinal arteries and veins (A/V) from the background, a task challenged by substantial variations in vessel calibers and the presence of low-contrast or ambiguous structures in fundus images, especially in ultra-wide field imaging where peripheral distortions and large-scale anatomical variability are pronounced. These factors often lead to fragmented semantic representations and topological inconsistencies in automated segmentation outputs. To address these limitations, we propose Ultra, a multi-granularity topological reasoning network designed for precise A/V segmentation. Ultra adopts a cascaded two-stage architecture: PriorNet generates coarse, multi-scale vascular priors that provide structural guidance, while RefineNet performs topology-aware segmentation refinement. To further enforce topological coherence, we propose the neighboring pixel connectivity regularization (NICER) layer, which selectively integrates local connectivity information predicted by the proposed connectivity prediction union (CPU) module. This connectivity is employed as auxiliary supervision through a pixel-wise local connectivity loss, reinforcing structural reasoning and promoting anatomically consistent vascular topology inference. Extensive experiments on ultra-wide field fundus imaging (UWF) datasets demonstrate that Ultra achieves state-of-the-art performance in A/V segmentation and topological preservation. Moreover, Ultra generalizes well to conventional color fundus photography (CFP) datasets, underscoring its robustness and broad applicability. Code is publicly available at: https://github.com/iMED-Lab/Ultra.

Among the more than 500 million tweets generated daily, a considerable number reflect public perspectives on global socio-political events such as the Russia–Ukraine War. Analyzing sentiment at this scale manually is time-consuming, inconsistent, and unsuitable for real-time decision-making. To overcome these limitations, this study presents a hybrid deep learning framework for automated sentiment analysis of Twitter discussions related to the Russia–Ukraine conflict. The approach begins with comprehensive Natural Language Processing (NLP) preprocessing, including tokenization, stopword removal, normalization, and lemmatization, followed by Exploratory Data Analysis (EDA) to analyze sentiment distribution, word frequency patterns, and textual trends. Context-aware semantic representations are generated using Distilled Robustly Optimized BERT Pretraining Approach (DistilRoBERTa), which offers efficient and lightweight transformer-based embeddings with reduced computational cost. To handle class imbalance and enhance minority class prediction, KMeans-Synthetic Minority Over Sampling Technique (SMOTE) is applied for synthetic data generation. For comparative evaluation, traditional classifiers such as Logistic Regression (LR), Support Vector Machine Classifier (SVC), and Random Forest Classifier (RFC) are utilized. The proposed model combines a Deep Neural Network (DNN) for effective feature learning with a Greedy Tree-based classifier (GTC) to improve classification performance and generalization capability. The resulting system, named “Semantic Distil Deep Tree (SDDT),” demonstrates superior performance over baseline models in terms of accuracy and F1-score. Overall, the proposed framework delivers a scalable, efficient, and dependable solution for real-time sentiment monitoring, aiding policymakers, researchers, and analysts in understanding public opinion on global conflicts.

Mobile reviews act as a key indicator of customer satisfaction, with over 90% of smartphone users referring to reviews before purchasing a device, and 72% of consumers indicating that positive feedback enhances their trust in a brand. However, manual examination of customer reviews and ratings is inefficient, susceptible to errors, and often fails to capture the subtle emotions present in unstructured text. To address these issues, this study proposes a framework based on Natural Language Processing (NLP) using an iPhone 14 dataset that includes user reviews, titles, and ratings. The workflow begins with NLP preprocessing and Exploratory Data Analysis (EDA) to clean, standardize, and visualize the data distribution. Next, Efficiently Learning an Encoder that Classifies Token Replacements Accurately (ELECTRA) is utilized for contextual feature extraction, enabling effective semantic representation of textual data. To handle class imbalance in review categories, K-Means Synthetic Minority Over-Sampling Technique (K-Means SMOTE) is applied to generate synthetic samples. Unlike existing approaches such as Adaptive Boosting Classifier (ABC) and Tao Tree Classifier (TTC), the proposed framework integrates an Extra Trees Classifier (ETC) to ensure more robust and scalable classification. The model predicts bivariate outputs Review Title and Rating thereby improving both sentiment understanding and rating consistency. By automating the analysis of reviews, the system provides valuable insights into customer satisfaction, product performance, and brand perception, ultimately supporting better decision-making and enhancing the overall customer experience.

Haul roads in surface mining are critical infrastructure directly influencing operational productivity, safety, and costs. However, these networks change frequently due to ongoing mining activities, making traditional mapping methods impractical for large-scale or rapidly evolving sites. Remote sensing imagery offers a scalable alternative, yet complex backgrounds, variable road widths, and spectral similarities between roads and surrounding surfaces make accurate extraction challenging. This study proposes HRM-Net, a hybrid transformer–CNN autoencoder framework for automated extraction of mine haul roads from remote sensing imagery. HRM-Net introduces inception-like patch embedding to capture local contextual information and employs a manifold-constrained hyper-connection strategy in the attention and fusion blocks to enhance information flow across the architecture. This hierarchical design enables progressive learning of discriminative semantic representations across multiple spatial resolutions, critical for road extraction in cluttered mining environments. Trained and evaluated on diverse mine sites, HRM-Net achieved 92.53% overall accuracy, 85.12% F1-score, 75.57% mIoU, 83.57% precision, and 86.94% recall, outperforming state-of-the-art transformer-based and CNN-based segmentation models. Furthermore, model interpretability was analyzed through linear probing and boundary alignment evaluations. Results demonstrate that discriminative features emerge at early network stages and are effectively preserved throughout the architecture, while boundary predictions exhibit superior consistency compared to existing approaches.

Deep learning is transforming microscopy, yet models often fail when applied to images from new instruments or acquisition settings. Conventional adversarial discriminative domain adaptation (ADDA) retrains entire networks, often disrupting learned semantic representations. Here, we overturn this paradigm by showing that adapting only the earliest convolutional layers, while freezing deeper layers, yields reliable transfer. Building on this principle, we introduce subnetwork image translation ADDA with automatic depth selection (SIT‐ADDA‐Auto), a self‐configuring framework that integrates shallow‐layer adversarial alignment with predictive uncertainty to automatically select adaptation depth without target labels. We demonstrate robustness via multimetric evaluation, blinded expert assessment, and uncertainty‐depth ablations. Across exposure and illumination shifts, cross‐instrument transfer, and multiple stains, SIT‐ADDA improves prediction fidelity and downstream segmentation over full‐encoder adaptation and nonadversarial baselines, with reduced drift of semantic features. Our results provide a design rule for label‐free adaptation in microscopy and a recipe for field settings; the code is publicly available.

In this study, we propose a new insider cyberthreat detection (ICD) framework called CMS-DDQN. The CMS-DDQN model is also integrated with ZTA in order to provide adaptive insider cyber defence. Additionally, the proposed framework combines multimodal behavioral analytics, semantic content representations, and RL – based decision-making in order to support a unified pipeline that is capable of performing detection, decision, and mitigation. The semantic embeddings extracted from file, email, and HTTP content using SBERT are compressed through self-supervised autoencoders to generate compact latent representations that are able to capture both behavioral semantics and contextual information. Accordingly, the agent learns adaptive access control policies within a custom ZT environment through four security actions: allow, limited access, escalation, and denial. Based on the experimental results, the evaluation on the CERT r6.2 dataset indicates that our CMS-DDQN framework is capable of achieving strong detection capability with 0.9947 accuracy, 0.9660 recall, 0.9328 precision, an F1-score of 0.9491, and an AUC of 0.9988. These findings indicate that the CMS-DDQN model has near-perfect discrimination between malicious and benign behaviors. The results also demonstrate that integrating semantic content awareness, self-supervised representation learning, and RL-based policy optimization significantly improves detection robustness and enables adaptive ZT enforcement.

This paper investigates the feasibility of using Large Language Models (LLMs) for the structural analysis of flow-based network data. This analysis is carried out in the presence of a structural difference between the multidimensional numerical space of IoT features and the symbolic space in which LLMs operate. The primary objective was the development of a formal framework that enables the controlled transformation of numerical data into linguistically analyzable semantic representations, without resorting to classification or machine learning mechanisms. We propose the Semantic Flow Encoding (SFE) mechanism, a deterministic method for robust discretization and behavioral abstraction that converts the numerical characteristics of Internet of Things (IoT) flows into structural semantic descriptions using the Canadian Institute for Cybersecurity Internet of Things Device Identification and Anomaly Detection (CIC IoT-DIAD) 2024 dataset. Through formal informational measures, it is demonstrated that the existence of an intrinsic structural difference between benign and DDoS traffic in the analyzed dataset. In the validation stage, we evaluated whether these informational differences are reflected at the level of linguistic abstraction through controlled inference experiments in IBM WatsonX. The present paper suggests that LLMs may support semantic auditing of distributional structure when guided by a formal encoding layer. In this manner, a reproducible framework for integrating numerical security data into language-model-based analysis is suggested.

The use of multimodal data is essential for the precise diagnosis and treatment of brain tumors. In this context, multimodal data encompass multisequence magnetic resonance imaging, computed tomography, positron emission tomography, histopathological images, molecular and genomic profiles, structured clinical variables, and radiological reports. With the rapid advancement of artificial intelligence, integrating these heterogeneous data sources has become acentral research direction for improving diagnostic accuracy, prognostic assessment, and therapeutic decision-making in neuro-oncology. However, substantial discrepancies exist across data modalities in terms of spatial resolution, semantic representation, and measurement scales, posing significant challenges for effective cross-modal integration. Multimodal large language models (MLLMs) enhance both interpretative and generative capabilities by jointly modeling visual, textual, and structured data, thereby offering aunified framework for addressing these challenges in brain tumor analysis. This review provides acomprehensive overview of MLLMs, covering their methodological foundations, representation learning strategies, and cross-modal alignment mechanisms. We further summarize their applications in both research and emerging clinical settings, including diagnosis support, prognosis prediction, treatment planning assistance, and radiology report generation. Finally, we discuss current limitations, such as data scarcity, interpretability constraints, and clinical deployment barriers, and outline future directions toward robust, explainable, and clinically translatable MLLM systems in neuro-oncology.