Table of Content

28 August 2025, Volume 11 Issue 8

Previous Issue   

Design of a Large Model Data Supervision System Based on Blockchain

2025, 11(8):  682. 

Asbtract ( )   PDF (2618KB) ( )  

References | Related Articles | Metrics

Large model (LM) has shown great potential in the fields of natural language processing, image and speech recognition, and has become a key force driving the technological revolution and social progress. However, the wide application of LM technology brings challenges such as data privacy risks, data compliance regulation, and data regulatory activation and intelligence.  This paper aims to explore how to utilize blockchain to design and construct an effective data regulatory system to promote its healthy development, in order to meet the challenges brought by the application of massive data to LM. This paper analyzes the trends and current status of the development of LM at home and abroad, and points out the main challenges to LM data regulation, including data privacy risks, data compliance, and the difficulty of effective supervision by regulators . A blockchainbased data regulation system design scheme is proposed to address these challenges, which realizes the fullcycle data regulation of LM data from the native metadata to the input of training until the posttraining feedback through four interconnected modules, namely, privacy protection, consensus algorithm, incentive mechanism, and smart contract. Finally, the application prospect of blockchain in LM data supervision is summarized, and the future trend of data supervision is outlooked.

Fake News Detection Model Based on Crossmodal Attention Mechanism and#br#  Weaksupervised Contrastive Learning#br#

2025, 11(8):  693. 

Asbtract ( )   PDF (1508KB) ( )  

References | Related Articles | Metrics

With the widespread popularization of the Internet and smart devices, social media has become a major platform for news dissemination. However, it also provides conditions for the widespread of fake news. In the current social media environment, fake news exists in multiple modalities such as text and images, while existing multimodal fake news detection techniques usually fail to fully explore the intrinsic connection between different modalities, which limits the overall performance of the detection model. To address this issue, this paper proposes a hybrid model of crossmodal attention mechanism and weaksupervised contrastive learning(CMAWSCL) for fake news detection. The model utilizes pretrained BERT and ViT models to extract text and image features respectively, and effectively fuses multimodal features through the crossmodal attention mechanism. At the same time, the model introduces weaksupervised contrast learning, which utilizes the prediction results of effective modalities as supervisory signals to guide the contrast learning process. This approach can effectively capture and utilize the complementary information between text and image, thus enhancing the performance and robustness of the model in multimodal environments. Simulation experiments show that the CMAWSCL performs well on the publicly available Weibo17 and Weibo21 datasets, with an average improvement of 1.17 percentage points in accuracy and 1.66 percentage points in F1 score compared to the current stateoftheart methods, which verifies its effectiveness and feasibility in coping with the task of multimodal fake news detection.

Encrypted Traffic Detection Method Based on Knowledge Distillation

2025, 11(8):  702. 

Asbtract ( )   PDF (2774KB) ( )  

References | Related Articles | Metrics

In recent years, with the rapid growth of Internet traffic, especially the popularity of encrypted communication, malicious traffic detection is facing a huge challenge, due to the limited resources and performance of mobile devices, which makes it more difficult to identify malicious behaviors in encrypted traffic on mobile. Therefore this paper proposes a knowledge distillation based encrypted traffic detection method. First, the traffic is transformed into images through visualization techniques; second, based on the ConvNeXt network architecture, the SK_SwiGLU_ConvNeXt network is constructed as the teacher network by introducing the SKNet attention mechanism and replacing the activation function GELU with SwiGLU; finally, the lightweight MobileNetV2 is selected as the student network and the use the teacher network to guide the student network training. The experimental results of this paper’s detection method on the publicly available dataset ISCX VPNNonVPN show that even in the resourceconstrained mobile device environment, the student network can improve the detection effect of the teacher model while reducing the model complexity, which proves that this method has efficient deployment potential on mobile devices.

A Privacy Budget Allocation Method Based on Differential #br# Privacy kmeans++#br#

2025, 11(8):  710. 

Asbtract ( )   PDF (1126KB) ( )  

References | Related Articles | Metrics

For the traditional differential privacy kmeans++ algorithm, uniform budget allocation by the equal division method cannot meet varying privacy needs. Meanwhile, binary division rapidly depletes the budget, leading to excessive noise later on, both impairing clustering performance. To solve this problem, a new privacy budget allocation method combining the arithmetic and equal allocation methods was proposed. For initial center selection, use an equal division budget allocation. For center updates, early stage uses arithmetic progression, later stage switches to equal division, both focused on minimal budget. This approach ensures substantial initial privacy budget for minimal cluster center distortion, and moderate budget depletion later to prevent excessive noise that could compromise clustering outcomes. A series of experiments based on real data show that, compared to the original kmeans++, the minimum error is only 0.09%. Compared to the equal distribution method and the binary method, the clustering accuracy is improved by up to 14.9% and 16.9% respectively. It can be seen that this method is significantly better than the equal division and the binary division, and can improve the usability and accuracy of clustering results to a certain extent.

Implicit Harmful Text Detection Technology Based on Knowledgeenhanced #br# Multitask Learning#br#

2025, 11(8):  718. 

Asbtract ( )   PDF (1578KB) ( )  

References | Related Articles | Metrics

A large number of harmful texts on the Internet adopt implicit and euphemistic expressions to evade detection by censorship systems. Most of the current work focuses on explicit harmful speech and cannot effectively detect implicit harmful text. This paper investigates the detection of implicit euphemistic harmful text in Chinese using a multitask learning approach, where euphemistic sentence recognition is used to assist harmful text detection. Firstly, methods for integrating euphemistic language vocabulary features are explored to enhance the model’s representation of implicit meanings. Subsequently, contrastive learning is applied to enhance latent semantic representations and extract common features from implicitly harmful discourse. Finally, a multitask learning framework is constructed by combining euphemistic sentence recognition tasks with harmful text detection tasks, aiming to improve the detection performance through shared multitask parameters and multifeature fusion loss functions. The experimental results demonstrate the effectiveness of the model in detecting implicit harmful text.

Optimization Method for Formal Analysis of Security Protocols #br# Based on Prior Path Selection#br#

2025, 11(8):  727. 

Asbtract ( )   PDF (2549KB) ( )  

References | Related Articles | Metrics

Formal analysis techniques can detect security vulnerabilities in security protocols, but when analyzing complex security protocols, the state space explosion problem often prevents the analysis from terminating. The fundamental cause of this issue is the exponential increase in the number of protocol states due to an excessive number of redundant nodes. To address this, the prior path selection method was proposed, which used the nodes of already searched paths to guide the selection of subsequent nodes, reducing the number of protocol states and effectively circumventing the state space explosion, thus improving efficiency. Further, by utilizing the Oracle interface of the Tamarin model checking tool, this method was applied to the analysis of security protocols, and comparative experiments were conducted on 5 protocols with 8 lemmas. The experimental results show that the Prior Path Selection method successfully provides analysis results for protocols where conventional path search is ineffective, mitigating the state space explosion problem.

Application Research of Differential Privacy Shuffle Model in Range Query

2025, 11(8):  736. 

Asbtract ( )   PDF (3395KB) ( )  

References | Related Articles | Metrics

Range queries are key indicators in data analysis under various scenarios. However, when dealing with individuallevel data, personal privacy issues will be involved. To address this problem, range query protocols that meet local differential privacy (LDP) have been proposed. These protocols enable data collectors to collect aggregated information about the population without relying on trusted third parties while protecting the privacy of each user. Nevertheless, the perturbation methods used in the existing range query protocols based on LDP have limitations, which restrict their effectiveness. In addition, these protocols usually exhibit poor estimation performance for small range intervals. In light of this, a Hierarchical Range Query protocol based on the differential privacy shuffling model (SHRQ) is proposed. Firstly, this paper extensively analyzes the variance of the perturbation methods in previous protocols. The SHRQ protocol selects the optimal perturbation method according to the number of nodes in each layer. Then, the SHRQ makes the most of the advantages of the shuffling model by leveraging prior knowledge from the previous round for multiple iterations, significantly improving the estimation accuracy of small range query intervals. Through extensive comparative experiments on both simulated and realworld datasets, it is demonstrated that after a few iterations, SHRQ reduces the estimation error for small ranges by an order of magnitude and for large ranges by half an order of magnitude compared to previous protocols.

A Privacy Protection Scheme for Blockchain Transaction Based on #br# Threshold Homomorphic Encryption#br#

2025, 11(8):  746. 

Asbtract ( )   PDF (1142KB) ( )  

References | Related Articles | Metrics

Blockchain is widely used because of its distributed processing, multiparty consensus, and immutable data. However, the open and transparent processing method will leak the privacy of users, and it is particularly important to use encryption technology to prevent the leakage of sensitive information. This paper proposes a privacy protection scheme for blockchain transactions based on threshold homomorphic encryption algorithm. Firstly, the confidentiality of sensitive transaction data is guaranteed by homomorphic encryption of the user’s account balance and transfer amount; Then, the corresponding transaction confirmation and transaction verification methods are designed. Finally, the security analysis and experimental verification of the proposed scheme are carried out, and the results show that the scheme has good stability and scalability, and is suitable for the general account model blockchain system.

Lightweighted Mutual Authentication and Key Agreement in V2N IoV

2025, 11(8):  753. 

Asbtract ( )   PDF (2403KB) ( )  

References | Related Articles | Metrics

Aiming at the scenario of vehicle secure access to application servers in the V2N (vehicle to network) environment, a Kerberos extension protocol is proposed based on the PUF (physical unclonable function). This protocol provides the twoway authentication and key agreement between the vehicle and the remoted application server and ensured the confidentiality and authentication of the V2N data transmission. The CRP (challenge response pair) generated by the PUF is used to replace the password in standard Kerberos to prevent the threats of key leakage caused by physical attacks such as intrusion, semiintrusion, sidechannel attacks, etc. The characteristics of Kerberos’s lightweighted twoway authentication protocol can overcome the defects of high calculation complexity and slow speed of the public key authentication algorithms, and effectively provide the secure data transmission between vehicles and application servers.

Research on Analysis and Detection Methods of Adversarial Crosssite #br# Scripting Attacks Based on LSTM and CNN#br#

2025, 11(8):  761. 

Asbtract ( )   PDF (1115KB) ( )  

References | Related Articles | Metrics

In recent years, machine learning and deep learning techniques have achieved significant success in detecting crosssite scripting (XSS) attacks. However, they still face challenges in defending adversarial attacks. To address this issue, this paper proposes an optimized method based on soft actorcritic (SAC) reinforcement learning combined with long shortterm memory (LSTM) and convolutional neural network (CNN). Firstly, adversarial samples are generated by leveraging the SAC and LSTMCNN detection model to simulate attacker strategies. These samples are then used for incremental training of the detection model, progressively narrowing the adversarial data generation space and improving the model’s robustness and detection accuracy. Experimental results show that the generated adversarial data achieves an evasion success rate of over 90% across multiple detection tools. After incremental training, the detection model’s defense capability against adversarial XSS attacks is significantly enhanced, with the evasion rate continuously decreasing.

Efficient Dynamic Multikey Fully Homomorphic Encryption Scheme #br# from LWE#br#

2025, 11(8):  768. 

Asbtract ( )   PDF (4153KB) ( )  

References | Related Articles | Metrics

The application of full homomorphic encryption in cloud computing effectively meets the user’s demand for “available but invisible” data over the cloud server. Aiming at the problems that the efficiency of multikey fully homomorphic encryption scheme needs to be optimized and the working mode applied to cloud computing is not reasonable, an efficient dynamic multikey fully homomorphic encryption scheme is proposed. On the one hand, the ciphertext extension algorithm of multikey homomorphic encryption scheme is optimized by introducing a pair of public keys and constructing new auxiliary ciphertexts. On the other hand, using a single user and the cloud server to complete the ciphertext extension operation, a new working mode of fully homomorphic encryption applied to cloud computing is proposed. Compared with the scheme of ICPADS meeting in 2023, our scheme reduces the computation overhead from O(n44) to O(n3k22), nk and noise expansion from O(m4γ) to O(mγ), making our scheme with smaller public parameters and more efficient. At the same time, the new working mode not only reduces the user’s high dependence on the server, but also reduces the computing overhead that the user needs to bear, and is more in line with the practical application. The scheme is proved to be INDCPA security and the difficulty can be reduced to the learning with error problem.