In this study, we address the specific challenges of Named Entity Recognition (NER) in Chinese legal contracts. Traditional Chinese NER systems struggle to balance the loss of important lexical information when using character-based segmentation and the high accuracy and extensive data requirements of word-level segmentation. To bridge this gap, we propose CW-BERT (Character-Word BERT), a novel framework that integrates external lexicon knowledge into BERT through two modules: Character Adapters (CA) and Word Adapters (WA). The Character Adapter enriches each character representation with weighted word-level features using a bilinear attention mechanism, while the Word Adapter updates word representations based on associated character sequences. These adapters are jointly trained with BERT and are injected into intermediate transformer layers, enabling a more expressive and lexically-aware contextual encoding. Extensive comparative experiments on Chinese legal contract datasets demonstrate that CW-BERT achieves better performance in legal contract NER tasks. Furthermore, the dynamic lexicon update mechanism not only proves valuable for enhancing legal knowledge extraction through transferable domain-specific vocabulary, but also demonstrates strong generalization potential in other fields such as medicine-validating its adaptability to diverse domain-specific linguistic features.

To address the limitations of existing trajectory privacy protection methods, which generally neglect users’ personalized privacy requirements and suffer from utility degradation under traditional noise-injection mechanisms, a personalized dynamic privacy trajectory protection based on the Laplace mechanism was proposed. First, a fuzzy sensitivity quantification model was developed to characterize the semantic sensitivity of each location point by jointly considering spatial contextual features and individual privacy preferences, thereby enabling personalized sensitivity assessment. Then, a dynamic privacy budget allocation strategy was designed to distribute privacy budgets according to the sensitivity levels of trajectory points, upon which the Laplace mechanism was applied to generate adaptive perturbations. This enabled stronger protection for highly sensitive regions while maintaining higher data fidelity for low-sensitivity areas. Experimental results show that, compared with representative differential privacy-based trajectory perturbation methods, the proposed approach achieved superior privacy preservation performance and significantly enhanced the utility of perturbed trajectory data.

Low Earth Orbit (LEO) mega-constellations promise seamless global service, yet satellites racing overhead force handovers every few minutes, disrupting service for billions of users worldwide. Current systems maintain exclusive single-path connections despite modern phased arrays’ ability to track multiple satellites simultaneously, accepting these disruptions as unavoidable and merely minimizing instead of eliminating them. We propose PMPS (Predictive Multi-Path Scheduling), designed to achieve handover-free satellite communications through proactive continuous multi-path service. PMPS eliminates the reactive single-path service interruption through three integrated designs: (i) introducing multiplicative composite weights combining visibility time, signal quality, and capacity factors orchestrates continuous traffic redistribution, ensuring gradual and seamless packet migration across all active satellites. (ii) proving optimal multi-satellite packet scheduling is NP-complete, then developing polynomial-time approximation algorithms achieves provably near-optimal performance within real-time constraints. (iii) eliminating obstruction-induced packet loss through anticipatory selective redundancy protects vulnerable connections. As satellites approach the horizon becoming vulnerable to obstruction, their declining weights have shifted most traffic to higher satellites. Redundantly transmitting only the minimal remaining packets through stable satellites ensures zero loss if obstruction occurs. Evaluation on Starlink and OneWeb constellations demonstrates PMPS enables handover-free LEO communications with minimal redundancy overhead.

To address the limitations of existing trajectory privacy protection methods, which generally neglect users’ personalized privacy requirements and suffer from utility degradation under traditional noise-injection mechanisms, a personalized dynamic trajectory privacy protection method based on the Laplace mechanism was proposed. First, a fuzzy sensitivity quantification model was developed to characterize the semantic sensitivity of each location point by jointly considering spatial contextual features and individual privacy preferences, thereby enabling personalized sensitivity assessment. Then, a dynamic privacy budget allocation strategy was designed to distribute privacy budgets according to the sensitivity levels of trajectory points, upon which the Laplace mechanism was applied to generate adaptive perturbations. This enabled stronger protection for highly sensitive regions while maintaining higher data fidelity for low-sensitivity areas. Experimental results show that, compared with representative differential privacy-based trajectory perturbation methods, the proposed method achieves superior privacy preservation performance and significantly enhances the utility of perturbed trajectory data.

Large-scale Low Earth Orbit (LEO) constellations often experience high node failure rates caused by dynamic environmental factors (random failures), such as satellite maneuvers, or by cyber or physical attacks on critical nodes (targeted attacks), which pose unique challenges for routing optimization. Traditional algorithms such as Dijkstra suffer from limited parallel scalability on GPUs due to irregular neighbor distributions, while deep learning methods lack generalization ability. To address these challenges, we propose TRACE (Topology-aware Routing via Adjacent-Constraint Encoding), a graph-constrained Transformer architecture equipped with a cascaded multi-head attention decoder for distributed dynamic routing in LEO domains. To improve algorithm throughput and resilience against random failures and targeted attacks, we further design NFD (Navigator-Follower Distillation), a self-distillation framework which enables each agent to learn routing policies from Monte Carlo episodes. Furthermore, a hierarchical distributed routing architecture is developed to extend the proposed method to multi-domain scenarios. Simulation results demonstrate that TRACE with NFD achieves near-optimal routing accuracy with controllable inter-domain errors, while significantly improving throughput compared with mainstream routing algorithms.

Contract review is a critical process to protect the rights and interests of the parties involved. However, this process is time-consuming, labor-intensive, and costly, especially when a contract faces multiple rounds of review. To accelerate the contract review and promote the completion of transactions, this paper introduces a novel benchmark of legal provision recommendation and conflict detection for contract auto-reviewing (ProvBench), which aims to recommend the legal provisions related to contract clauses and detect possible legal conflicts. Specifically, we construct the first Legal Provision Recommendation Dataset: ProvData, which covers 8 common contract types. In addition, we conduct extensive experiments to evaluate ProvBench on various state-of-the-art models. Experimental results validate the feasibility of ProvBench and demonstrate the effectiveness of ProvData. Finally, we identify potential challenges in the ProvBench and advocate for further investigation.

Aspect Sentiment Triple Extraction (ASTE) is an emerging task in sentiment analysis that aims to extract triplets consisting of aspect terms, opinion terms, and sentiment polarity from review texts. Previous span-based methods often struggle with accurately identifying the boundaries of aspect and opinion terms, especially when multiple word spans appear in a sentence. This limitation arises from their reliance on a single, simplistic approach to constructing contextual features. To address these challenges, we propose Multi-Perspective Boundary Enhancement Network (MPBE). The network captures rich contextual features by adopting a dual-encoder mechanism and constructs multiple channels to further enhance these features. Specifically, we introduce enhanced semantic and syntactic information in two channels, while the third channel transforms the features using discrete fourier transform. In addition, we design a dual-graph cross fusion module to fuse features from different channels for more efficient information interaction and integration. Finally, by statistically analyzing the length distribution of aspect and opinion terms, a candidate length-based decoding strategy is proposed to achieve more accurate decoding. In experiments, the proposed MPBE model achieved excellent results on four benchmark datasets (14Lap, 14Res, 15Res, 16Res), with F1 scores of 62.32

Aspect Sentiment Triplet Extraction (ASTE) is a method for extracting aspect terms, opinion terms, and their corresponding sentiment polarities from a given sentence. Most of the existing studies use joint extraction methods to extract the triplets directly in a unified framework. However, most joint extraction methods only consider the semantic and syntactic dependency information of the sentence. Due to a lack of sentiment information and positional information, they are unable to accurately and completely express the aspect and opinion in the sentence. In order to solve the above problems, we introduce a Multiple-level Enhanced Graph Convolutional Network (MEGCN) for ASTE, which utilizes sentiment scores and sentiment polarity nodes alongside syntactic dependency information. This approach not only enriches contextual understanding by integrating sentiment data but also improves positional analysis of aspect and opinion terms through polarity nodes. Moreover, our dual-aware fusion module, combining semantic with sentiment-enhanced syntactic features through a biaffine attention mechanism and matrix construction, enables a deeper representation of aspect sentiment triplets. Our model demonstrates superior performance over existing methods on two widely recognized ASTE datasets.

Network Functions Virtualization (NFV) and Service Orchestration have significantly streamlined the deployment and management of satellite networks which are characterized by their large-scale, multi-domain structure, with each domain managed by its own NFV MANO. Furthermore, the deployment of Service Function Chains (SFCs) frequently spans numerous domains, necessitating collaboration among the NFV MANOs across these domains. However, cross-domain collaboration poses several serious challenges, including autonomous domain managing, limited inter-domain information sharing, and insufficient global cross-domain SFC deployment optimizing. Therefore, cooperative cross-domain deployment of SFCs in satellite networks includes two fundamental sub-problems: inter-domain VNF assignment and intra-domain VNF embedding. For the former we propose a novel Auction-based Proximal Policy Optimization (APPO) method to encode the crossed domain of SFC. For the latter we employ the Long Short-Term Memory (LSTM)-based Actor-Critic method. Additionally, we design a adjacent domain resource abstraction method to simplify the cross-domain physical resource view for reinforcement learning. Finally, we conduct extensive experiments to demonstrate the superior performance of this method in real cross-domain SFC deployment scenarios.

Anomaly detection in Unmanned Surface Vehicle (USV) is a critical task for ensuring maritime safety. Traditional approaches often rely on unimodal data, such as GPS trajectories, which fail to fully capture the dynamic behaviors in complex marine environments, leading to sub-optimal detection results. To address above challenges, in this paper, we propose the Dual-Branch and Dual-View Multimodal Learning (DBDVML) framework which initially incorporates visual information with GPS trajectories to enhance anomaly detection in USV. Firstly, we design the Visual Dual-Branch Dual-View Module (VDBDV), which extracts both holistic and localized semantics for comprehensive utilization of dual-view visual environmental contexts. Secondly, the Multimodal Semantic Fusion Module (MSF) is developed, employing both intra and inter modal relation exploration to effectively enhance the semantic expression, thereby achieving more precise anomaly detection. Extensive experiments on a real-world USV dataset demonstrate the effectiveness of our DBDVML framework, significantly outperforming existing methods in terms of anomaly detection accuracy. This work highlights the potential of utilizing multimodal data, specifically, vision and GPS modalities, for robust anomaly detection in complex marine environments. Moreover, its principles can also be extended to other multimodal domains.The code is publicly available at https://github.com/Tmygithub/DBDVM.

Low Earth Orbit (LEO) mega-constellations enable the Internet of Things (IoT) industry to realize the vision of integrated space-air-ground-sea communication networks under B5G and 6G. However, due to the high complexity of O(|V|log|V|+|E|) of Dijkstra’s algorithm, existing LEO mega-constellations suffer from excessive routing reconvergence time under frequent topology changes caused by satellite-ground station link handovers and network failures. To this end, we propose MLDR, a Manhattan-like topology-and Low ISL Delay-based Routing scheme with an ultra-low routing complexity of O(|V|4) while maintaining near-optimal routing. Firstly, for any source satellite, MLDR divides the Manhattan-like topology of LEO mega-constellations into four non-interfering Minimum Hop (MH) areas. Secondly, MLDR concurrently and non-repeatedly computes MH paths for all destinations within each MH area and installs routing tables. Thirdly, MLDR incrementally computes and updates the MH paths by more optimal MH detour paths, which select other neighboring satellites as relaying nodes. Finally, by conducting extensive simulations on real-world LEO mega-constellations, our MLDR outperforms all state-of-the-art schemes by remarkably reducing reconvergence time and achieving the highest routing optimality.

Link prediction is a network analysis model used to discover missing links or future relationships that may appear, which has been widely used in many real network systems to predict the potential relationship between two individuals. However, link prediction can also be used by attackers to identify sensitive links that users are unwilling to expose, which makes only removing sensitive links from the original network ineffective and leads to the disclosure of privacy. In this paper, we propose a target link protection mechanism via artificial bee colony algorithm based on random walk (RABC), which can defend the link prediction attacks based on Resource Allocation (RA) metric effectively.To enhance the local search ability of RABC, the random walk algorithm is combined with the original artificial bee colony algorithm. Then, we compare our method with other existing methods, which shows that RABC has higher efficiency while ensuring the effectiveness. Finally, extensive experiments on real social networks are conducted to demonstrate the good performance of RABC on protecting sensitive links from being detected successfully by link prediction model. Furthermore, the perturbed networks generated by RABC is transferable to defend against other link prediction attacks.

The adoption of SDN/NFV opens avenues for efficient network slicing deployment and cost control. However, the dynamic cost reduction brought by deployment location optimization is not suitable for all scenarios. To further reduce the cost, we recommend a sharing strategy in NFV. In this paper, we introduce a two-layer VNF multiplexing architecture, named PSMA, which guarantees both efficient VNF sharing operations and secure slicing during multiplexing. Leveraging the SDN/NFV features, the proposed scheme splits data processing and key management and establish secure connections using SDN’s programmable routing. The provided framework integrates comprehensive life-cycle management and key delivery mechanisms. The article substantiates its availability through extensive simulations of VNF reuse on a randomly generated network with Virtual Network Requests (VNR). The empirical results indicate a significant cost reduction of 5% to 10%, particularly pronounced in scenarios involving a substantial number of short-lived and transitional VNFs.

Firstly,the current situation of satellite remote sensing,communication,and computing systems was systematically analyzed,and the real-time remote sensing dilemma existing in the traditional satellite remote sensing systems and the satellite remote sensing communication integrated system architecture was deeply analyzed.Secondly,a multi-satellite and satellite-ground coordination inte-grated system architecture for satellite remote sensing,communication,and computation was put forward,and the key components and the working theories modes of the architecture were introduced in detail.Thirdly,the advantages of this architecture were demonstrated and the feasibility of key technologies was systematically clarified.Finally,the simulation and analysis method based on this architec-ture,which could reach minute-level remote sensing tasks with high timeliness and high potential application value compared with the traditional mode,was elaborated.

A power distribution system is usually composed of a number of distribution substations and distribution lines, where the latter are fragile and often suffer from faults, leading to serious local power shortages. For all distribution substations, it is complex and costly to diagnose and locate faulty input lines from the multiple input lines of distribution substations. This paper reports a Petri net strategy to diagnose the faulty input lines to prevent local power shortages. The structure of a supervisor is proposed to detect and locate the faulty input lines by supervising the variations of power loads that are stored in a battery energy storage system. The stored power loads can be uninterruptedly distributed to the output lines during the fault diagnosis processes. Moreover, there are other auxiliary distribution substations that can urgently supply power loads to a distribution substation once the faulty input lines of the distribution substation are diagnosed. The local power shortages of the distribution substation are prevented during the fault diagnosis processes. In addition, the structures of power distribution systems and their control systems are formally defined by using Petri nets. The derivation processes of fault detection and location strategies are verified by the mathematical methods of Petri nets. Finally, a case study is presented.

Social networks especially the social communities facilitate the rapid and rich social activities of all individuals in the world. However, advanced community detection brings serious privacy disclosure (e.g., whether two important and sensitive individuals belong to the same community?) to us. For instance, online plainclothes policemen can be regarded as marginal community users who are often in the same community initially and in need of penetrating into as many as possible different communities to collect illegal evidence of network criminals, but adversarial community inference which can maliciously disclose the sensitive user relationships within a target community will expose their privacy and lead to task failure. Thus, privacy protection for the marginal community users becomes an urgent issue which is still open so far. In this work, we aim to study the community privacy protection for target marginal individuals of a community against multiple adversarial community detection (ACD) attacks. First, we define the marginal community user hiding problem and propose a marginal user pair selection strategy. Second, to enhance the privacy effectiveness of conventional methods, we propose a deep graph learning approach to maximally find the minimum link perturbation cost. Finally, we conduct various community detection attacks on many real social graphs, and the experimental results show that our method can more effectively hide the marginal-sensitive user pairs than baselines.

The emergence of generative models has revolutionized the field of remote sensing (RS) image generation. Despite generating high-quality images, existing methods are limited in relying mainly on text control conditions, and thus do not always generate images accurately and stably. In this paper, we propose CRS-Diff, a new RS generative framework specifically tailored for RS image generation, leveraging the inherent advantages of diffusion models while integrating more advanced control mechanisms. Specifically, CRS-Diff can simultaneously support text-condition, metadata-condition, and image-condition control inputs, thus enabling more precise control to refine the generation process. To effectively integrate multiple condition control information, we introduce a new conditional control mechanism to achieve multi-scale feature fusion, thus enhancing the guiding effect of control conditions. To our knowledge, CRS-Diff is the first multiple-condition controllable RS generative model. Experimental results in single-condition and multiple-condition cases have demonstrated the superior ability of our CRS-Diff to generate RS images both quantitatively and qualitatively compared with previous methods. Additionally, our CRS-Diff can serve as a data engine that generates high-quality training data for downstream tasks, e.g., road extraction. The code is available at https://github.com/Sonettoo/CRS-Diff.

Distributed computing systems based on cloud-edge-device have been successfully serving thousands of applications and have become mainstream,characterized by a wide audience,high user experience requirements,and high security expectations.However,in recent years,frequent attacks on cloud-edge-device systems have resulted in serious security risks and significant economic losses for users.The defense mechanisms of cloud-edge-device systems have been found to operate independently,leading to major differences in the ability to resist risks,which makes it difficult to eliminate global security risks.Security measures of the pre-configuration and post-event remedy type have been found to hardly meet the security needs of high-load and high real-time in cloud-edge-device collaborative networks.The root cause has been identified as the hierarchical architecture of cloud-edge-device systems being separate,and no effective collaborative defense system has been formed,leading to problems such as difficult cross-domain security governance,poor real-time performance,and difficult collaborative consistency evaluation.Firstly,a cloud-edge-device collaborative security architecture that integrates real-time perception,dynamic decision,and proactive defense was proposed.The cloud-edge-device security collaboration capability was established through rigorous theoretical consistency proofs.Secondly,a collaborative security assessment model based on the order parameter was put forward to achieve consistency of risk perception,decision,and defense.In this way,the efficient and low-cost collaborative protection of security risks could be realized to maximize systemic security benefits.Finally,the consistency theory and assessment method were verified through simulations.The results show that the proposed collaborative security system and consistency evaluation model are correct and effective.

The Proactive Recovery (PR) mechanism in Software-Defined Networking (SDN) provides good failure recovery resilience for the Beyond Fifth-Generation/Sixth-Generation (B5G/6G) delay-sensitive applications. However, PR’s fixed single backup path policy for any flow and fine-grained backup forwarding rule configuration poses severe challenges for post-recovery congestion management and limited Ternary Content Addressable Memory (TCAM) space in SDN switches. To this end, we propose K-backup, a load-and TCAM-aware multi-backup fast failure recovery scheme for SDNs. Firstly, K-backup formulates and solves the congestion-aware multi-backup path planning problem for various failure scenarios, exploiting the inherent load diversity of multi-backup paths to minimize the post-recovery maximum link utilization. Secondly, K-backup aggregates flows sharing the same backup-path-weight pair on a link into a cascading table of Fast-Failover and SELECT groups. Meanwhile, each outputted backup path is labeled, and a corresponding label-matching flow table is configured for each intermediate switch to aggregate all flows on that path. Thirdly, K-backup dynamically adjusts the backup path update period based on the network load stabilization to reduce unnecessary controller overhead. Compared with state-of-the-art, K-backup achieves the lowest controller overhead, the best load balancing performance, the near-fewest TCAM space usage, and the near-shortest recovery time.