Short-text classification remains an important research topic in natural language processing, yet it continues to face challenges caused by semantic sparsity and limited annotated data. Recent studies have attempted to combine graph neural networks with contrastive learning to alleviate these issues. However, existing methods often fail to fully exploit the latent clustering structure of unlabeled corpora and make insufficient use of scarce prior annotations, thereby limiting model performance. To address these limitations, we propose a Clustering-Guided Graph Contrastive Learning framework for Short-Text Classification (CG-STC).Specifically, we construct a corpus-level heterogeneous graph consisting of a word graph, a linguistic-information graph, and an entity-annotation graph, and employ a shared-parameter encoder to generate augmented views. Within the contrastive learning framework, original and augmented representations are aligned to refine text embedding distributions. K-means clustering is then applied to generate pseudo-labels for unlabeled texts, while a centroid contrastive loss is introduced to regularize the clustering process and enhance semantic consistency. The generated pseudo-labels are further incorporated into a supervised contrastive learning stage as supervisory signals.Experimental results on multiple public benchmark datasets demonstrate that the proposed method achieves competitive performance compared with conventional approaches, and on certain datasets performs comparably to, or better than, contemporary large language models. The proposed framework provides an effective solution for short-text classification under semantic sparsity and limited annotation settings.

In cross-enterprise supply chain collaboration scenarios, transaction data such as transaction amounts and cross-enterprise exchange rates all possess rational number properties. However, existing approaches to processing rational numbers in secure computing involve scaling by the same multiple, which may result in loss of rational number precision. Any loss of precision or simplistic conversion of rational numbers to integers may pose risks of privacy leaks and erroneous judgments. To address the aforementioned issues, this paper studies the application of secure computation based on the intersection and union of rational number sets in cross-enterprise supply chain collaboration. this paper transforms rational numbers into matrices and designs secure computation protocols for set intersection and union under semi-honest and malicious models using the Elgamal homomorphic encryption algorithm. The security of the protocol under the malicious model is proven using an ideal-practical example. Finally, this paper analyzes the efficiency of existing protocols versus the proposed protocol, and proves its practicality.

In the process of open source software development, teamwork has become a mainstream trend. However, developers generally tend to communicate with acquaintances, and it is difficult to find high-quality unfamiliar developers, which brings a series of negative effects. To some extent, developers’ development efficiency and enthusiasm are negatively affected, and open source projects are easy to fall into the “homogenization trap”, and even die early. At the same time, the open source community faces the risk of isolation and rigidity. Therefore, developer recommendation is an important task for improving the efficiency of developers, promoting the rapid iteration of technology, and continuing to inject vitality into the open source community. To this end, we design an LLM-and-Edge enhanced HGT model (LEHGT), and investigate the developer recommendation task on the datasets constructed from the GitHub community. The core idea is to evaluate the relationships between developers and repositories in text and use the evaluation feature to enhance the HGT-based recommender. Experimental results show that the proposed model performs significantly better than comparative methods.

As the foundation of aquatic ecosystem food chains, zooplankton play a pivotal role in maintaining ecological balance and promoting biodiversity. However, zooplankton recognition still relies on expert knowledge and manual analysis of printed materials and microscopic images, resulting in low efficiency. Moreover, owing to a lack of structured domain knowledge, the accuracy of automated recognition models such as ZooSCAN and question-answering models is limited, particularly when handling zooplankton species with similar morphologies. To bridge these gaps, this study proposes a multimodal domain ontology construction method based on OWL, focusing on dominant zooplankton species in Shandong offshore area, China. This approach enables the formal representation of zooplankton taxonomic knowledge and further constructs a multimodal zooplankton knowledge graph by embedding identification feature instances. This work provides a knowledge base and technical support for high-accuracy automatic recognition models, question-answering systems, biodiversity analysis, and ecological conservation of regional zooplankton.

With the rapid development of the Internet of Vehicles (IoV), achieving trustworthy vehicle position verification while preserving location privacy has become a key requirement in intelligent traffic supervision scenarios such as defense control zones and urban restricted-access areas. Existing privacy-preserving schemes have difficulty simultaneously supporting accurate determination of complex-shaped prohibited areas and efficient computation, and still face malicious attacks such as interference with verification procedures, tampering with communication processes, and privacy inference when determining the positional relationship between vehicles and prohibited areas. To address these issues, this paper proposes an efficient privacy-preserving position verification (PPPV) scheme based on secure multi-party computation (MPC). The scheme supports arbitrary polygonal prohibited areas, including convex, concave, and self-intersecting polygons, thereby improving its applicability in complex IoV supervision scenarios. Based on an improved cross-product determination method, this paper constructs an efficient PPPV protocol under the semi-honest model, achieving near-plaintext computational efficiency while protecting the privacy of both vehicle locations and area boundaries. To resist malicious attacks, this paper further combines Paillier homomorphic encryption, the cut-and-choose method, and zero-knowledge proof to construct a secure PPPV protocol under the malicious model, which can effectively prevent protocol deviations, result tampering, and inference attacks. This paper also conducts formal security proof based on the real/ideal model paradigm, and evaluates the performance of the scheme through benchmark experiments and attack experiments. Experimental results show that the scheme achieves a good balance among efficiency, applicability, and security, providing a deployable trustworthy position verification mechanism for next-generation IoV intelligent supervision applications.

Aiming at the problem of confidential determination of the positional relationship between automatic guided vehicle (AGV) and elliptical forbidden area in intelligent warehousing, this paper transforms the problem into the problem of confidential determination of the positional relationship between a point and an ellipse and proposes a determination scheme based on a secure multi-party computation, which utilizes Paillier encryption algorithm to substitute the point coordinates into the elliptical general-form equations and realizes the confidential determination of the positional relationship by the sign of the computation results. Firstly, the confidential determination protocol of the positional relationship between point and ellipse under the semi-honest model is designed, based on which the potential malicious attack behaviors are analyzed in depth, including malicious behaviors such as the malicious adversary forging the calculation results or terminating the protocol. In response to these malicious behaviors, a confidentiality determination procedure was created using a malicious model and encryption techniques including zero-knowledge proofs and the cut-and-choose. The efficiency of the strategy suggested in this study has greatly increased after an analysis of its correctness, likelihood of successful attacks by malevolent adversaries, and efficiency.

Cross-Domain Sequential Recommendation (CDSR) seeks to improve user preference modeling by transferring knowledge from multiple domains. Despite the progress made in CDSR, most existing methods rely on overlapping users or items to establish cross-domain correlations-a requirement that rarely holds in real-world settings. The advent of large language models (LLM) and model-merging techniques appears to overcome this limitation by unifying multi-domain data without explicit overlaps. Yet, our empirical study shows that naively training an LLM on combined domains-or simply merging several domain-specific LLMs-often degrades performance relative to a model trained solely on the target domain. To address these challenges, we first experimentally investigate the cause of suboptimal performance in LLM-based cross-domain recommendation and model merging. Building on these insights, we introduce WeaveRec, which cross-trains multiple LoRA modules with source and target domain data in a weaving fashion, and fuses them via model merging. WeaveRec can be extended to multi-source domain scenarios and notably does not introduce additional inference-time cost in terms of latency or memory. Furthermore, we provide a theoretical guarantee that WeaveRec can reduce the upper bound of the expected error in the target domain. Extensive experiments on single-source, multi-source, and cross-platform cross-domain recommendation scenarios validate that WeaveRec effectively mitigates performance degradation and consistently outperforms baseline approaches in real-world recommendation tasks.

The Intrusion Detection System (IDS) is a security mechanism developed to observe network traffic and recognize suspicious or malicious activities. Clustering algorithms are often incorporated into IDS; however, conventional clustering-based methods face notable drawbacks, including poor scalability in handling high-dimensional datasets and a strong dependence of outcomes on initial conditions. To overcome the performance limitations of existing methods, this study proposes a novel quantum-inspired clustering algorithm that relies on a similarity coefficient-based quantum genetic algorithm (SC-QGA) and an improved quantum artificial bee colony algorithm hybrid K-means (IQABC-K). First, the SC-QGA algorithm is constructed based on quantum computing and integrates similarity coefficient theory to strengthen genetic diversity and feature extraction capabilities. For the subsequent clustering phase, the process based on the IQABC-K algorithm is enhanced with the core improvement of adaptive rotation gate and movement exploitation strategies to balance the exploration capabilities of global search and the exploitation capabilities of local search. Simultaneously, the acceleration of convergence toward the global optimum and a reduction in computational complexity are facilitated by means of the global optimum bootstrap strategy and a linear population reduction strategy. Through experimental evaluation with multiple algorithms and diverse performance metrics, the proposed algorithm confirms reliable accuracy on three datasets: KDD CUP99, NSL_KDD, and UNSW_NB15, achieving accuracy of 98.57%, 98.81%, and 98.32%, respectively. These results affirm its potential as an effective solution for practical clustering applications.

With the increasing use of electronic devices, online handwriting verification has become crucial for biometricsbased identity authentication. Traditional methods, which rely on content-dependent verification of the writer's name, are vulnerable to forgery. This paper introduces a content-independent handwriting authentication system, Ph-Wri, designed for commodity smartphones. The core innovation is a multi-path attention feature fusion network that combines both static features (image of the handwritten text) and dynamic features (time-dependent properties during writing), to abstract the handwriting style instead of specific content for recognition, enabling robust user authentication. To extract handwriting style from dynamic writing features, we propose a polarity-aware attention strategy during training. This strategy incorporates Style Channel Attention (SCA) to capture direction-sensitive stylistic features, and Trajectory Spatial Attention (TSA) to highlight key handwriting trajectory regions. In the fine-tuning stage, the Correlation-Aware Attention (CAA) module models inter-channel structural correlations, mitigating the influence of content and enhancing style-consistent representations. By linking content-independent handwriting style to user identity, the system achieves accurate authentication. Extensive experiments on both the self-built CIEHD dataset and the public BiosecurID dataset demonstrate exceptional performance, achieving a 99% Verification Accuracy on CIEHD. Compared to state-of-theart methods that utilize only static or dynamic data, Ph-Wri significantly reduces the Equal Error Rate, showcasing the effectiveness and practicality of the proposed approach.

Smart-home consumer electronics (CE) sharded blockchains face a configuration dilemma: node-resource information is scattered across home gateways, while shard configuration requires global coordination. Centralized optimization exposes local resource states, making it difficult to balance real-time performance and security-related risk exposure. To address this problem, this paper proposes FedShard-DES, a FedPareto-driven double-elite-selection framework for federated multi-objective optimization of sharded blockchain configuration. The proposed model jointly optimizes shard failure probability, transaction throughput, decentralization entropy, and consensus latency. In FedShard-DES, each CE client locally performs double-elite-selection evolutionary search and uploads only its Pareto archive instead of raw node-resource data. The coordinator aggregates local archives through non-dominated sorting, Minkowski-distance-based truncation, and feedback of representative Pareto solutions to support distributed Pareto optimization. Experiments on the smart-home sharded blockchain model and WFG benchmark suites show that FedShard-DES obtains a favorable security-oriented trade-off compared with centralized and federated baselines. The representative and default-budget validations further show that FedShard-DES maintains favorable trade-off performance under comparable federated settings.

Data aggregation can integrate scattered and isolated vehicle data into useful information for real-time traffic analysis, road condition prediction, or intelligent decision-making. In the Internet of Vehicles (IoV) scenario, secure data aggregation ensures the privacy and security of original vehicle data (such as location, speed, and driving status), which is of great significance for complying with user privacy protection regulations and preventing sensitive information leakage. Based on the threshold NTRU encryption algorithm and a novel encoding method, this paper proposes a multi-party secure data aggregation protocol suitable for IoV. Furthermore, to address potential cheating behaviors by malicious participants, an anti-cheating protocol is designed to detect and prevent the submission of false data in critical steps. Both protocols can compute statistical metrics (total count, sum, average, maximum, minimum, median, variance, and standard deviation) for multi-party vehicle data in a single run and support partial result output on demand. Simulation-based proofs demonstrate that the protocols can resist n-1 collusion attacks. Performance analysis and experimental tests show that the protocols outperform existing schemes in computational and communication efficiency, and are particularly adaptable to the high dynamics and real-time requirements of IoV. This study provides an efficient and feasible solution to the problem of privacy-preserving data aggregation in IoV.

Federated learning (FL) enables privacy-preserving model training for medical consumer electronics (CE) devices, such as wearable sensors, smartphones, and home health monitoring terminals. However, large-scale heterogeneous device resources and Non-IID medical data may bias FL resource scheduling toward resource-rich or data-rich clients, resulting in cross-device service inconsistency, long aggregation delay, and high energy consumption. To address this problem, this paper proposes a medical-CE-oriented FL resource allocation framework for cross-device service consistency. Under a fixed per-round client participation budget, the framework jointly optimizes client selection, local processor frequency, and transmission power. The resource allocation task is formulated as a large-scale four-objective optimization problem, where global model accuracy, model accuracy variance, aggregation delay, and energy consumption are optimized simultaneously, with model accuracy variance used to characterize cross-device service consistency. To solve this problem, a large-scale multi-objective evolutionary algorithm based on fuzzy grouping (LSMOEA/FG) is developed to identify convergence-sensitive and diversity-sensitive variables and optimize them cooperatively. Experimental results show that FL based on LSMOEA/FG improves global model accuracy by 5.83% on average and reduces model accuracy variance, aggregation delay, and energy consumption by 17.22%, 23.01%, and 8.02%, respectively, compared with five state-of-the-art evolutionary methods.

The Internet of Medical Things (IoMT) refers to the number of interconnected medical objects and applications that are essential in contemporary healthcare. Nevertheless, its dependence on interrelated ecosystems subjects it to high risks of cybersecurity, including data breaches, privacy breaches, and malicious attacks that threaten patient safety and system integrity. This paper resolves these issues by presenting a new and enhanced intrusion detection model, BiGRU/RBWK, which is a combination of Bidirectional Gated Recurrent Units (BiGRU) with a Refined Black-winged Kite (RBWK) optimization algorithm. The RBWK algorithm improves on the BiGRU model by optimizing its hyperparameters, resulting in faster convergence and better classification performance. The model uses a powerful preprocessing pipeline, which includes Recursive Feature Elimination (RFE) that is used in conjunction with Support Vector Machines (SVM), which helps to curb the overfitting and lessen the computational cost. The experimental findings with two public datasets, namely WUSTL-EHMS-2020 and ECU-IoHT, indicate that BiGRU/RBWK model displays higher accuracy (95.6% and 93.8% respectively), precision, recall, F1-score, and AUC-ROC than all other currently available methods, including Random Forest, LSTM, CNN, Autoencoder, and CNN-LSTM hybrid models. The suggested framework has a high discrimination capability and can be easily used in real-time applications in the resource-constrained IoMT settings.

The Industrial Internet of Things (IIoT) enhances production efficiency through interfactory resource sharing and intelligent human-machine interaction (HMI) in flexible manufacturing. This study addresses large-scale multiobjective resource scheduling optimization in the complex IIoT environments to improve production efficiency and resource scheduling under growing customization and order demands. We develop a multiobjective IIoT scheduling model under the assumptions of task priority and resource constraints, optimizing makespan (MS), logistics time (LT), energy consumption (EC), tardiness time (TT), production cost (PC), and carbon emission (CE). To improve resource scheduling performance, a fuzzy decision-enhanced dual-directed sampling-assisted large-scale multiobjective optimization algorithm (LMOEA-FDDS) is proposed. The algorithm employs a dual-directed sampling (DS) method, uses two different types of search directions to guide the population evolution, and combines a complementary environmental selection strategy based on angle penalty distance (APD), effectively improving the algorithm's search efficiency and performance. Experimental results show that LMOEA-FDDS achieves 30%-70% higher HV values than the state-of-the-art algorithms, demonstrating its effectiveness in solving complex IIoT scheduling problems.

Educational recommendation systems have been a fundamental component for alleviating learning disorientation in self-paced learning. While existing studies mainly leverage cognitive theories to guide learning motivation modeling, they critically overlook the role of social influences. Through empirical analysis, we identify social homophily as an additional driver of learning behaviors, i.e., learners tend to adopt resources validated by their social cohort. However, two challenges impede effective social homophily modeling: (1) the absence and sparsity of predefined social relations in online education, and (2) the deep entanglement of social homophily with cognitive homophily in behavioral data. To tackle these challenges, we propose a graph-based framework EdGCL that explicitly disentangles social homophily and cognitive homophily. EdGCL infers implicit social relations from learners' social behaviors and encodes them via a graph transformer, generating social-view representations. Simultaneously, it constructs a heterogeneous learning graph to model cognitive homophily, which is enhanced by a type-aware aggregator and cognitive diagnosis loss. To ensure the semantic distinctiveness of dual-view homophily modeling, a cross-view contrastive disentanglement mechanism is designed to pull intra-view representations closer while pushing inter-view representations away. Evaluation on two real-world educational datasets demonstrates the superior recommendation performance of EdGCL, highlighting the necessity of dual homophily modeling for understanding the motivations behind learning behaviors.

The remaining oil distribution plays a vital role in enhanced oil recovery (EOR), which directly guides the de velopment of oilfields. However, efficiently predicting the remaining oil distribution is a challenge due to the complex reservoir fluid distribution after water flooding. As a popular method for remaining oil studies, reservoir numerical simulation is frequently confronted with challenges such as high computational complexity and long matching time. This paper proposes a reservoir cellular automata (RCA) to predict the remaining oil distribution during water flooding. We divide the oil reservoir into a grid space, which is mapped to a cellular space. Each grid is a cellular node in the cellular space, and its state is characterized by oil saturation. The eight nodes geograph ically adjacent to a node are its neighbor nodes, forming a Moore-type neighborhood. We design lightweight deduction rules that transform the remaining oil dynamic evolution into nodes state iterative updates. Based on the pressure difference, Darcy's law calculates the volume change of two-phase fluid, which is combined with porosity to update the node state. Then, the pressure after the fluid flow is updated by the material balance equation. In addition, a data supplement method is also presented to initialize the RCA. On a sand body in north ern China, we apply the RCA to predict the remaining oil distribution based on reservoir data. Experiments show that RCA reduces deduction time by 87 % compared to numerical simulation while achieving an MRE of less than 10 %.

Smart contract is a computerized protocol for informationally disseminating, validating, or enforcing a contract when specific conditions are met. The similarity calculation of the graph is used to compare the similarity between different smart contracts to detect potential malicious contracts or plagiarism. At present, most of the secure multi-party calculations of graph structure similarity are designed under the semi-honest model, which cannot resist malicious enemies. In this paper, a new coding method based on prime number is proposed and a semi-honest computing graph similarity protocol is designed by using ElGamal Threshold encryption algorithm. Then we analyze the possible malicious attack behavior in the semi-honest model protocol and design the secret computational graph similarity protocol with the help of zero-knowledge proof cryptography tool. Finally, the real/ideal model paradigm is used to prove that the protocol is secure under the malicious model, and the efficiency of the protocol is proved by comparing with the existing schemes.

Open source communities have a wealth of digital talents, who are urgently needed by various industries under the digitalization process of the entire society. However, barriers exist between digital talents in open source communities and employers. On one hand, open source contributors wonder whether their expertise matches the requirements of specific jobs; On the other hand, developers working on small open source projects are less likely to get recognition from employers, compared with those contributing to well-known projects. To bridge this gap, we propose a new task, matching digital talents and job titles in open source communities, which measures the matching degrees between digital talents with open source experience and job titles requiring digital skills. To solve the task, we construct a heterogeneous information network connecting open source communities and job markets, and propose a semi-supervised network alignment model to augment the connectivity of the network. Then we employ a graph neural network to learn the representations of the digital talents and the job titles from the augmented network, based on which we measure the matching degrees between them. Experimental results demonstrate that our method achieves improvements of at least 5.34, 3.52, 2.37, 2.93, and 8.21 in accuracy, precision, recall, F1, and AUC compared to other possible solutions.

The rapid advancement of modern wireless communication systems has driven an urgent need for complex optimization involving high-dimensional objectives, multiple goals, and non-convex search spaces in multiple-input and multiple-output (MIMO) antenna arrays. However, conventional and quantum-inspired algorithms encounter several limitations, such as an inadequate balance between exploration and exploitation and premature convergence. To address these challenges, this study proposes a hybrid Quantum-Inspired Artificial Bee Colony Algorithm with Dynamic Programming (DP-QABC) for improved MIMO beamforming optimization. The core innovation lies in a hybrid architecture that integrates four quantum-inspired evolution strategies with a diversity-aware dynamic programming controller. The quantum-inspired components, including qubit encoding, quantum rotation gates, entanglement operations, and quantum tunneling, are designed to adaptively expand the global search space. Additionally, a simulated annealing-based acceptance criterion and a cluster-based local exploitation strategy are incorporated to enhance convergence quality and guide the search through complex solution landscapes. The proposed DP-QABC is evaluated comprehensively on both benchmark functions and practical MIMO beamforming scenarios. Experimental results demonstrate that it achieves at least a 40% improvement in search capability and a 8.4%-17.3% reduction in computation time compared to the existing QABC approach. In beamforming applications, an overall performance improvement of approximately 17.1% over QABC is observed, while effectively maintaining high directional gain, interference suppression and low sidelobe levels. These results validate DP-QABC as a robust and efficient solution for complex optimization problems in wireless communications.