Traditional public opinion diffusion models generally assume interactions between individuals as binary pair-wise effects, which struggle to capture the higher-order complexities of multi-group interactions in social networks—such as group discussions in WeChat and topic reposting on Weibo. Moreover, these models fail to adequately depict the nonlinear trust accumulation mechanisms and individual heterogeneity inherent in the diffusion process. Therefore, the paper proposes a hypergraph-based Hyper-S2IR model for disseminating public opinion. The “Goebbels effect” is operationalized by leveraging the hypergraph structure: a susceptible node’s risk of infection is proportional to its hyperdegree, mathematically representing the cumulative exposure to information from multiple sources within different hyperedges. Our model introduces two types of communicators (HI1 and HI2) with different motivations and capabilities, thereby systematically depicting the inherent heterogeneity of the communication group. Through theoretical derivation, we derive a novel basic reproduction number R0 that explicitly incorporates the hyperdegree distribution of the hypergraph. This R0 provides a threshold for dissemination dynamics: When R0 > 1, the public opinion will continue to spread and converge to a stable public opinion prevalence equilibrium point; when R0 < 1, the public opinion will gradually disappear. Critically, the expression for R0 reveals how higher-order group interactions, encoded in the hyperdegree, fundamentally alter the spreading threshold compared to traditional pairwise networks. Numerical simulations verify the theoretical conclusions and demonstrate that the hypergraph structure significantly accelerates the spread and expands the scale of public opinion compared to traditional network structures. This work provides theoretical support and a quantitative basis for analyzing public opinion dissemination mechanisms and formulating intervention strategies.

Abstract Existing Transformer-based models for fine-grained image classification often struggle to explicitly model the granular hierarchical relationships among multi-scale features and perform adaptive fusion. This limitation leads to the insufficient utilization of discriminative features. Inspired by granular computing theory, this paper proposes a Multi-Granularity Collaborative Network (MGCN-FGIC) that integrates core operations—including granulation, enhancement, alignment, and fusion—into a deep learning framework. Specifically, the network reorganizes and enhances backbone features via a Granule-aware Enhancer and achieves adaptive collaborative fusion of cross-granularity information using a Granule Fusion Neck. Furthermore, a Dynamic Hard Example Mining mechanism is designed to trigger multi-granularity supervision based on sample difficulty, thereby improving discriminative ability for challenging examples. The proposed method achieves classification accuracies of 92.95

The partitioning mechanism of fuzzy hypergraphs often relies on empirical rules, lacking a clear, traceable path, which limits the model’s credibility and generalizability. Fuzzy Random Forest can enhance the interpretability of decision path generation. Based on this, a fuzzy hypergraph construction method called FRF-FHG is proposed. First, construct a fuzzy random forest based on the data, extract the complete branch paths from the root node to the leaf node for each tree in the FRF, define the sample set satisfying all feature constraints along the path as a fuzzy hyperedge, and achieve an effective mapping from the rule space to the hypergraph structure; Secondly, to address the issues of redundant branch paths, the explosive increase in fuzzy hyperedges, and “densification” caused by the inherent characteristics of the FRF model’s branching mechanism, a fuzzy hyperedge reduction strategy is further designed. This strategy optimizes the hypergraph topology while precisely preserving the core relational information among samples. Finally, through example calculations and experimental analysis, FRF-FHG is applied in classification applications to deeply analyze the underlying mechanisms of model judgment errors, providing theoretical support for performance optimization; Four typical datasets from different fields were selected for comparative and ablation experiments, and the results showed that the classification performance of FRF-FHG can reach a robust and excellent level, with outstanding performance in core classification indicators on multiple actual datasets.

To address the current issue that existing steganographic techniques are highly dependent on the quality of carrier images and have insufficient adaptability to damaged images, we propose an MAE-based inpainting-steganography framework, which realizes steganography and inpainting for 2D color damaged images through multi-module collaboration. The encoder divides the original 244 × 244 images into non-overlapping patches and performs random masking to generate their feature sequences. The steganographic point prediction module locates suitable steganographic points by analyzing feature fluctuations during the inpainting process. The steganography module embeds information by introducing feature offsets at specific positions in the feature sequences according to the suitable steganographic points. The decoder completes the inpainting of masked regions to obtain stego-inpainted images. And the extraction module recovers the secret information through feature alignment. Experimental results show that with a feature fluctuation amplitude threshold of 0

High-dimensional mixed data often lack a unified semantic representation for continuous and discrete attributes, which hinders mixed-attribute similarity modeling and can result in unstable reducts and overfitting in existing neighborhood rough set (NRS) methods. To address this issue, we propose IF-EMD-SPA, an attribute reduction method for NRS grounded in Information Flow theory. Unlike conventional NRS methods that rely on discretization or a single reduction criterion, IF-EMD-SPA first establishes a unified representation framework for heterogeneous attributes based on classifications and an Information Channel Core. It then integrates Earth Mover's Distance (EMD) and Set Pair Analysis (SPA) to define a similarity metric for mixed attributes. In addition, a three-stage greedy reduction strategy is designed under the dual constraints of dependency preservation and structural error, consisting of dependency-driven forward selection, similarity-driven structure completion, and backward redundancy removal. Experiments on five UCI benchmark datasets and two high-dimensional gene expression datasets show that IF-EMD-SPA achieves average accuracies of 93.5% (k-Nearest Neighbors, KNN), 93.9% (Support Vector Machine, SVM), and 90.8% (Classification and Regression Trees, CART), with SVM achieving the best results on all seven datasets. Under CART, it reaches 100% accuracy on Wine and WPBC, improving performance by up to 37.5 percentage points over comparison methods.

The Social Internet of Things (SIoT) combines social networks and the Internet of Things, enabling closer interaction between devices, users, and services. However, this interaction brings risks of trust attacks. These trust attacks not only affect the stability of SIoT systems but also threaten personal privacy and data security. This paper provides a decade-long review of SIoT trust attack research. First, it outlines the SIoT architecture, social relationship types, concept of trust, and trust management processes. It maps seven attacks—bad mouthing attack (BMA), ballot stuffing attack (BSA), self-promoting attack (SPA), discriminatory attack (DA), whitewashing attack (WWA), on-off attack (OOA), and opportunistic service attack (OSA)—clarifying their mechanisms and traits. Next, we synthesize the literature on SIoT trust models, enumerate which attack types they address, and classify defense strategies. It then conducts simulation-based comparative experiments on trust attacks to reveal their impact on node trust and transaction processing, compares attack capabilities along disruption speed, attack strength, and stealthiness, and summarizes attack surfaces with corresponding defense recommendations to better guide the design of SIoT trust management schemes. Finally, we identify open challenges and future research directions, to support the development of new trust management models better equipped to address evolving trust attacks.

In recent years, a large number of works have introduced Convolutional Neural Networks (CNNs) into image steganography, which transform traditional steganography methods such as hand-crafted features and prior knowledge design into steganography methods that neural networks autonomically learn information embedding. However, due to the inherent complexity of digital images, issues of invisibility and security persist when using CNN models for information embedding. In this paper, we propose Curriculum Learning Progressive Steganophy Network (CLPSTNet). The network consists of multiple progressive multi-scale convolutional modules that integrate Inception structures and dilated convolutions. The module contains multiple branching pathways, starting from a smaller convolutional kernel and dilatation rate, extracting the basic, local feature information from the feature map, and gradually expanding to the convolution with a larger convolutional kernel and dilatation rate for perceiving the feature information of a larger receptive field, so as to realize the multi-scale feature extraction from shallow to deep, and from fine to coarse, allowing the shallow secret information features to be refined in different fusion stages. The experimental results show that the proposed CLPSTNet not only has high PSNR , SSIM metrics and decoding accuracy on three large public datasets, ALASKA2, VOC2012 and ImageNet, but also the steganographic images generated by CLPSTNet have low steganalysis scores.You can find our code at \href{https://github.com/chaos-boops/CLPSTNet}{https://github.com/chaos-boops/CLPSTNet}.

As a high-dimensional extension of ordinary graphs,hypergraphs can more flexibly reflect high-order complex relationships between nodes.Hypergraph clustering aims to discover complex high-order correlations in powerful hypergraph structures.In response to challenges faced by current hypergraph clustering algorithms,such as extremely high complexity,unstable clustering results,and the tendency to fall into local optima,a multi-stage hypergraph clustering algorithm denoted as MSHC is proposed based on the idea of hypergraph partitioning.This algorithm divides the hypergraph clustering process into three stages:hypergraph reduction,hypergraph initial clustering,and optimization migration.In the first stage,a fast reduction method that preserves the hypergraph structure is proposed to reduce the complexity of subsequent algorithms.In the second stage,a similarity measurement method between hypergraph nodes based on set pair analysis theory is introduced,and hierarchical clustering algorithm is applied for initial clustering.Four different cluster merging strategies are employed to increase the diversity of clustering schemes.In the final stage,the genetic algorithm is applied to obtain the optimal hypergraph clustering scheme.Comparative experiments are conducted with two traditional hypergraph clustering algorithms on three data sets with different sizes.Experimental results show that the hypergraph modularity index of the MSHC algorithm is improved by 0.079 and 0.077on the Songs_genres and Papers_keywords datasets respectively,and is only reduced by 0.006 on the Movies_genres dataset.

Existing network intrusion detection models effectively capture relationships between nodes and extract key features. However, they often struggle to accurately represent node characteristics, particularly in modeling the spatiotemporal dynamics and topological structures with sufficient granularity. To address these limitations, we propose the dynamic interaction multi-channel graph convolutional network (DIMK-GCN), which integrates three key components: a spatiotemporal feature weighting module, an interactive graph feature fusion module, and a temporal feature learning module. The spatiotemporal feature weighting module constructs a dynamic graph structure that incorporates both nodes and edges, leveraging self-attention mechanisms to enhance critical feature representations. The interactive graph feature fusion module employs graph attention networks (GATs) to refine node relationships while integrating a multi-channel graph convolutional network (GCN) to extract multi-perspective features, thereby enhancing model depth and robustness. The temporal feature learning module utilizes gated recurrent units (GRUs) to effectively capture long-term dependencies and address challenges posed by non-stationary time series data. Experimental results on the CIC-IDS2017, CIC-IDS2018, and Edge-IIoTSet datasets demonstrate that DIMK-GCN significantly outperforms existing models in key performance metrics, including detection accuracy, recall, and F1-score. Notably, on the Edge-IIoTSet dataset, DIMK-GCN achieves an accuracy of 97.31%, verifying its effectiveness and robustness in detecting various types of network attacks.

Aiming at the problems of poor quality of steganographic images and slow network convergence of image steganography models based on deep learning, this paper proposes a Steganography Curriculum Learning training strategy (STCL) for deep learning image steganography models. So that only easy images are selected for training when the model has poor fitting ability at the initial stage, and gradually expand to more difficult images, the strategy includes a difficulty evaluation strategy based on the teacher model and an knee point-based training scheduling strategy. Firstly, multiple teacher models are trained, and the consistency of the quality of steganographic images under multiple teacher models is used as the difficulty score to construct the training subsets from easy to difficult. Secondly, a training control strategy based on knee points is proposed to reduce the possibility of overfitting on small training sets and accelerate the training process. Experimental results on three large public datasets, ALASKA2, VOC2012 and ImageNet, show that the proposed image steganography scheme is able to improve the model performance under multiple algorithmic frameworks, which not only has a high PSNR, SSIM score, and decoding accuracy, but also the steganographic images generated by the model under the training of the STCL strategy have a low steganography analysis scores. You can find our code at \href{https://github.com/chaos-boops/STCL}{https://github.com/chaos-boops/STCL}.

For deep learning-based image steganography frameworks, in order to ensure the invisibility and recoverability of the information embedding, the loss function usually contains several losses such as embedding loss, recovery loss and steganalysis loss. In previous research works, fixed loss weights are usually chosen for training optimization, and this setting is not linked to the importance of the steganography task itself and the training process. In this paper, we propose a Two-stage Curriculum Learning loss scheduler (TSCL) for balancing multinomial losses in deep learning image steganography algorithms. TSCL consists of two phases: a priori curriculum control and loss dynamics control. The first phase firstly focuses the model on learning the information embedding of the original image by controlling the loss weights in the multi-party adversarial training; secondly, it makes the model shift its learning focus to improving the decoding accuracy; and finally, it makes the model learn to generate a steganographic image that is resistant to steganalysis. In the second stage, the learning speed of each training task is evaluated by calculating the loss drop of the before and after iteration rounds to balance the learning of each task. Experimental results on three large public datasets, ALASKA2, VOC2012 and ImageNet, show that the proposed TSCL strategy improves the quality of steganography, decoding accuracy and security.

The surface texture of objects in industrial scenes is complex and diverse, and the characteristics of surface defects are often very similar to the surrounding environment and texture background, so it is difficult to accurately detect the defect area. However, when deep learning technology is used to detect complex texture surface defects, the detection accuracy is not high, due to the lack of large-scale pixel-level label datasets. Therefore, a defect detection model Siamese-RCNet for complex texture surface with a small number of annotations is proposed. The Cascade R-CNN target detection network is used as the basic framework, making full use of unlabeled image feature information, and fusing the nonlinear relationship learning ability of Siamese network and the feature extraction ability of the Res2Net backbone network to more effectively capture the subtle features of complex texture surface defects. The image difference measurement method is used to calculate the similarity between different images, and the attention module is constructed to weight the feature map of the feature extraction pyramid, so that the model can focus more on the defect area and suppress the influence of complex background texture area, so as to improve the accuracy of detection. To verify the effectiveness of the Siamese-RCNet model, a series of experiments were carried out on the DAGM2007 dataset of weakly supervised learning texture surface defects for industrial optical inspection. The results show that even if only 20% of the labeled datasets are used, the mAP@0.5 of the Siamese-RCNet model can still reach 96.9%. Compared with the traditional Cascade R-CNN and Faster R-CNN target detection networks, the Siamese-RCNet model has high accuracy, can reduce the workload of manual labeling, and provides strong support for practical applications.

With the continuous development of autonomous driving, real-time target detection has become increasingly critical in autonomous driving systems. However, traditional target detection algorithms usually require huge computational resources, limiting their application on embedded autonomous driving platforms. To address this challenge, a lightweight single-stage target detection algorithm is proposed YOLOdrive. The inverted residual structure, linear bottleneck layer, and depth-separable convolution in MobileNetv2 are utilized to improve the YOLOv8 backbone network, while the spatial channel reconstructed convolution is used to improve the C2f module of YOLOv8, and the convolution of YOLOv8 neck and detection head is replaced by the depth-separable convolution. Experimental results verify that the average accuracy of YOLOdrive algorithm on MS COCO2017 data set and VOC2007 data set is improved compared with the baseline model YOLOv8-Nano. The amount of model parameters has been reduced by more than 50%, and the amount of computation in the model has been reduced by more than 70%. The algorithm drastically reduces the amount of parameters and computational complexity of the network, improves the operational efficiency, saves the storage space of the network, and maintains a high-detection performance.

Aiming at the fact that the moving target defense game model fails to accurately portray attack and defense gains, resulting in bias in attack and defense games and the inability to select effective defense strategies, we construct the moving target three-way evolutionary game defense model (MTTEGDM). Firstly, the model is defined and analyzed theoretically under the premise of uncertainty and irrationality. Then, combined with the three-way decisions, the attack intention is introduced into the target network loss calculation, and a dynamic weight adjustment algorithm based on the three-way decisions is proposed to accurately characterize the attack and defense gains from a multi-attribute perspective. Finally, the evolutionary game model is used to analyze the evolution trend of the multi-stage defense strategy, so as to carry out feasible and effective defense behavior. The simulation results show that the model can accurately predict the optimal defense strategy of moving targets in different stages. Through a Monte Carlo simulation experiment, the proposed algorithm is compared with the traditional evolutionary game model, and the effectiveness and security of the proposed algorithm are verified.

Credit risk assessment is an important process in bank financial risk management. Traditional machine-learning methods cannot solve the problem of data islands and the high error rate of two-way decisions, which is not conducive to banks’ accurate credit risk assessment of users. To this end, this paper establishes a federated three-way decision incremental naive Bayes bank user credit risk assessment model (FTwNB) that supports asymmetric encryption, uses federated learning to break down data barriers between banks, and uses asymmetric encryption to protect data security for federated processes. At the same time, the model combines the three-way decision methods to realize the three-way classification of user credit (good, bad and delayed judgment), so as to avoid the loss of bank interests caused by the forced division of uncertain users. In addition, the model also incorporates incremental learning steps to eliminate training samples with poor data quality to further improve the model performance. This paper takes German Credit data and Default of Credit Card Clients data as examples to conduct simulation experiments. The result shows that the performance of the FTwNB model has been greatly improved, which verifies that it has good credit risk assessment capabilities.

There are many problems in Social Internet of Things(IoTs), such as complex topology information, different degree of association between nodes and overlapping communities. The idea of set pair information grain computing and clustering is introduced to solve the above problems so as to accurately describe the similarity between nodes and fully explore the multi-community structure. A Set Pair Three-Way Overlapping Community Discovery Algorithm for Weighted Social Internet of Things (WSIoT-SPTOCD) is proposed. In the local network structure, which fully considers the topological information between nodes, the set pair connection degree is used to analyze the identity, difference and reverse of neighbor nodes. The similarity degree of different neighbor nodes is defined from network edge weight and node degree, and the similarity measurement method of set pair between nodes based on the local information structure is proposed. According to the number of nodes' neighbors and the connection degree of adjacent edges, the clustering intensity of nodes is defined, and an improved algorithm for initial value selection of k-means is proposed. The nodes are allocated according to the set pair similarity between nodes and different communities. Three-way community structures composed of a positive domain, boundary domain and negative domain are generated iteratively. Next, the overlapping node set is generated according to the calculation results of community node membership. Finally, experiments are carried out on artificial networks and real networks. The results show that WSIoT-SPTOCD performs well in terms of standardized mutual information, overlapping community modularity and F1.

入侵检测是目前网络安全防护的一个重要环节,由于传统入侵检测模型时间长、学习能力弱,因此提出了一种基于增量主成分分析方法(Incremental Principal Component Analysis,IPCA)和卷积神经网络(Convolutional Neural Networks,CNN)结合的方法,融合增量的思想能够使模型不断更新并拥有持续的学习能力.首先用IPCA方法对数据集进行降维处理,该算法将样本分批传入模型进行训练以不断更新特征基以及均值;最后找出主成分特征子集,再用CNN对其进行分类训练.实验采用 KDD-CUP99 以及 UNSW-NB15 作为实验的数据集进行对比.研究结果表明,IPCA-CNN 模型的准确率、F1 值和误报率分别达到了99.7％、99.3％和 0.2％;同传统机器学习入侵检测算法相比有所提升,验证了IPCA-CNN模型的有效性.

As a kind of special graph of structured data, a hypergraph can intuitively describe not only the higher-order relation and complex connection mode between nodes but also the implicit relation between nodes. Aiming at the limitation of traditional distance measurement in high-dimensional data, a new method of hypergraph construction based on set pair theory is proposed in this paper. By means of dividing the relationship between data attributes, the set pair connection degree between samples is calculated, and the set pair distance between samples is obtained. Then, on the basis of set pair distance, the combination technique of k-nearest neighbor and ε radius is used to construct a hypergraph, and high-dimensional expression and hypergraph clustering are demonstrated experimentally. By performing experiments on different datasets on the Kaggle open-source dataset platform, the comparison of cluster purity, the Rand coefficient, and normalized mutual information are shown to demonstrate that this distance measurement method is more effective in high-dimensional expression and exhibits a more significant performance improvement in spectral clustering.

Aiming at the problems of the dynamic increase in data in real life and that the naive Bayes (NB) classifier only accepts or rejects the sample processing results, resulting in a high error rate when dealing with uncertain data, this paper combines three-way decision and incremental learning, and a new three-way incremental naive Bayes classifier (3WD-INB) is proposed. First, the NB classifier is established, and the distribution fitting is carried out according to the minimum residual sum of squares (RSS) for continuous data, so that 3WD-INB can process both discrete data and continuous data, then carry out an incremental learning operation, select the samples with higher data quality according to the confidence of the samples in the incremental training set for incremental learning, solve the problem of data dynamics and filter the poor samples. Then we construct the 3WD-INB classifier and determine the classification rules of the positive, negative and boundary domains of the 3WD-INB classifier, so that the three-way classification of samples can be realized and better decisions can be made when dealing with uncertain data. Finally, five discrete data and five continuous data are selected for comparative experimental analysis with traditional classification methods. The results show that 3WD-INB has high accuracy and recall rate on different types of datasets, and the classification performance is also relatively stable.

In recent years, significant progress has been made in the application of federated learning (FL) in various aspects of cyberspace security, such as intrusion detection, privacy protection, and anomaly detection. However, the robustness of federated learning in the face of malicious attacks (such us adversarial attacks, backdoor attacks, and poisoning attacks) is weak, and the unfair allocation of resources leads to slow convergence and inefficient communication efficiency regarding FL models. Additionally, the scarcity of malicious samples during FL model training and the heterogeneity of data result in a lack of personalization in FL models. These challenges pose significant obstacles to the application of federated learning in the field of cyberspace security. To address these issues, the introduction of meta-learning into federated learning has been proposed, resulting in the development of federated meta-learning models. These models aim to train personalized models for each client, reducing performance discrepancies across different clients and enhancing model fairness. In order to advance research on federated meta-learning and its applications in the field of cyberspace security, this paper first introduces the algorithms of federated meta-learning. Based on different usage principles, these algorithms are categorized into client-level personalization algorithms, network algorithms, prediction algorithms, and recommendation algorithms, and are thoroughly presented and analyzed. Subsequently, the paper divides current cyberspace security issues in the network domain into three branches: information content security, network security, and information system security. For each branch, the application research methods and achievements of federated meta-learning are elucidated and compared, highlighting the advantages and disadvantages of federated meta-learning in addressing different cyberspace security issues. Finally, the paper concludes with an outlook on the deep application of federated meta-learning in the field of cyberspace security.