Bài báo quốc tế
Lĩnh vực nổi bật
Trí tuệ nhân tạo
922
Internet vạn vật
23
An toàn bảo mật
11
Blockchain
3
Tìm kiếm theo
Năm xuất bản
Lĩnh vực bài báo
Kho tri thức
/
Bài báo quốc tế
Sắp xếp theo
Liên quan
AnoResLSTM: A Hybrid Deep Learning Framework for Real-Time Cheating Detection in Online Exams
Bùi Quốc HuyIn this paper, we investigate cheating behaviors in computer-based examinations and propose an effective detection approach based on candidate behavioral analysis. The dataset was collected through simulated exam sessions and subsequently annotated frame by frame to support model training. To extract relevant features of the head, eyes, and mouth, we employed the Face Mesh framework from Mediapipe. We further introduce AnoResLSTM, a hybrid model combining a ResNet module with an AttLSTM network, designed to enhance the representation and analysis of sequential behavioral patterns. Experimental results demonstrate that our proposed method achieves high accuracy in detecting cheating while maintaining efficient processing time.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
RL-HCR: A Reinforcement Learning Based Adaptive Leader Selection Framework for Energy-Efficient WSNs
Trần Huy LongAs critical infrastructure for IoT applications, wireless sensor networks must minimize energy consumption to extend the operational lifetime of the entire network. Energy-efficiency protocols, such as LEACH and PEGASIS, have been developed over many years to face challenges. However, static cluster head selection mechanisms in these original operations produce suboptimal energy efficiency and limited adaptability, motivating a move toward dynamic leader selection. PEGABC, a PEGASIS variant that employs a metaheuristic algorithm for leader selection, reduces some local inefficiencies but incurs substantial re-clustering overhead and optimises only short-term gains. To overcome these limitations, we propose RL-HCR(Reinforcement Learning–Guided Hierarchical Chain Routing), a lightweight RL framework that dynamically determines when to recluster while preserving the chain-based energy efficiency of PEG ABC. The proposed RL-HCR achieves low inference cost and avoids wasteful reclustering. Simulation results under standard scenarios show that RL-HCR extends 20% lifetime compared to PEG ABC, while reducing total energy consumption. These findings suggest that RL-HCL is a viable and scalable approach for adaptive, energy-efficient routing in wireless sensor networks.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
A Dual-Path approach for Time Series Anomaly Detection in Building Environmental Sensors
Nguyễn Chí Minh HiếuAnomaly detection within environmental time series data plays a crucial role in modern monitoring systems, yet it continues to pose challenges due to the inherently complex and nonlinear nature of sensor-generated signals. This study proposes a dual-path approach for time series anomaly detection that combines the expressive capabilities of deep learning with the transparency of classic machine learning techniques. The approach integrates a bidirectional Long Short-term memory (LSTM) autoencoder for extracting temporal features with density-based outlier detection algorithms, specifically Local Outlier Factor (LOF) and Isolation Forest. This methodology effectively models time dependent patterns while maintaining a balance between interpretability and computational cost. The proposed approach shows significant improvements compared to standalone deep learning and conventional statistical approaches, across various evaluation metrics through extensive testing on indoor environmental sensor datasets. The results analyze the different impacts of components on the anomaly detection process: Isolation Forest (49.08%), Reconstruction Error (39.27%), and LOF (11.65%). Using synthetic data with differing noise intensities improved the model’s resilience across diverse anomaly categories—point, contextual, and collective-achieving detection rates above 86%. These findings
highlight the approach’s practical value in real-world environmental monitoring by balancing high accuracy and interpretability.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Refurbished Smartphones in the Circular Economy: Insights From Environmental and Consumer Perspectives
Nguyễn Đình SơnThis study investigates key factors shaping consumer acceptance of refurbished smartphones as a means to promote sustainable consumption and reduce electronic waste. Drawing on environmental psychology and trust theory, it examines the effects of environmental awareness, environmental concern, and social influence on consumer preference and purchase intention, while assessing the moderating roles of trust in products and retailers and the mediating role of preference. Survey data from 824 Vietnamese consumers were analyzed using structural equation modeling. The findings highlight environmental concern as the strongest driver of preference, with trust enhancing several relationships. The study offers practical insights for marketers and extends sustainability research in emerging markets.
Năm:2025
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản: Circular Economy and Sustainability
Analyzing Effects of News on Manipulated Stock Price using Large Language Model and Statistical Tests: Evidence in Vietnam Market
Trần Quốc KhánhIn emerging markets such as Vietnam, where retail investors dominate and regulatory enforcement remains limited, financial news plays a pivotal role in shaping stock price dynamics. This study examines the relationship between financial news sentiment and stock price manipulation using the case of the FLC Group. PhoBERT, a pre-trained Vietnamese large language model, is employed to classify financial news into positive, neutral, and negative categories. The resulting daily sentiment scores are aggregated and analyzed alongside stock price data from 2018 to 2023 through Pearson correlation, Granger causality, and Threshold Vector Autoregression tests. The empirical results reveal significant correlations and causal relationships between sentiment and stock prices, though their strength weakens after 2022, reflecting shifts in investor behavior and market transparency following regulatory events. The nonlinear analysis further demonstrates that sentiment–price interactions are regime-dependent, with mean reversion in pessimistic periods and momentum effects in optimistic ones. These findings highlight the vital role of news sentiment in influencing and potentially manipulating prices in emerging markets, providing insights for investors, policymakers, and regulators.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Adaptive federated learning with k-Means++ for rare-class IoT intrusion detection
Huỳnh Trọng ThưaIoT network intrusion detection faces the dual challenges of non-IID client data and severe class imbalance, under which rare yet security-critical attacks are systematically under-detected. This work addresses that gap by introducing FedMADE Weighting, a lightweight two-tier federated learning framework that explicitly optimizes rare-class detection. Clients train standard MLPs with inverse-frequency loss to highlight rare patterns, while the server performs rarity-aware dynamic aggregation within profile-based client clusters (formed via k-means++), rewarding updates that demonstrate rare-class competence and penalizing model drift. On CICIoT2023 (7.8M flows, 34 classes), with 100 clients and Dirichlet α = 0.5 partitions, FedMADE achieves 89.5% rare-class recall (RRA) at 94.9% macro-F1, and surpasses 90% macro-F1 within ≈15 rounds, indicating reduced communication compared with standard averaging. Ablation results show that client reweighting is the dominant driver (75.1% RRA alone), while the server’s rarity-aware weighting amplifies it to 89.5% without compromising macro-F1. Representative baselines (e.g., FedAvg, FedNova, DL-BiLSTM) report strong aggregate F1 on CICIoT2023 but do not evaluate rare-class recall, whereas FedMADE substantially improves the detection of rare attacks under realistic non-IID partitions with modest client cost and practical aggregation.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
ANALYZING THE IMPACT OF IT CAPABILITY ON BUSINESS PERFORMANCE OF SMES IN THE CONTEXT OF DIGITAL TRANSFORMATION IN HCMC
Nguyễn Văn SáuThis study examines the impact of e-commerce on the economic growth of small and medium-sized enterprises (SMEs) in Ho Chi Minh City, Vietnam. Grounded in the Technology-Organization-Environment (TOE) framework and Resource-Based View (RBV) theory, the research investigates five key factors: IT infrastructure, e-commerce capability, digital marketing, business data analytics, and employee IT skills. Using a quantitative approach, data were collected from 222 SMEs operating in e-commerce through structured questionnaires with a 5-point Likert scale during August-October 2025. The data were analyzed using SPSS 26.0, employing Cronbach's Alpha reliability testing, exploratory factor analysis (EFA), Pearson correlation analysis, and multiple linear regression. Results demonstrate that all five factors positively and significantly influence SME economic growth, with the model explaining 67.2% of the variance (R² = 0.672, F = 88.446, p < 0.001). IT infrastructure emerged as the strongest predictor (β = 0.301), followed by digital marketing (β = 0.277), e-commerce capability (β = 0.249), business data analytics (β = 0.242), and employee IT skills (β = 0.117). All hypotheses were supported with high statistical significance. The findings provide valuable insights for SME managers to prioritize investments in digital infrastructure and marketing strategies, while policymakers should develop supportive policies for digital transformation. This research contributes to the literature by providing empirical evidence of specific e-commerce factors driving economic growth in SMEs within an emerging market context.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Hybrid Federated Learning with TabTransformer and FedMADE-GSA for IoT Intrusion Detection
Huỳnh Trọng ThưaFederated Learning (FL) has emerged as a revolutionary paradigm for privacy-preserving machine learning across distributed Internet of Things (IoT) environments. However, contemporary federated learningbased intrusion detection systems encounter challenges such as non-IID data distributions, class imbalance, limited communication resources, and the progression of cyber threats. This paper introduces a novel hybrid federated learning framework that integrates three critical components: TabTransformer for optimal tabular data representation, FedMADE Weighting for adaptive client aggregation employing Class Probability Matrices (CPM) and client clustering through DBSCAN via a systematic
optimization process, and Gradient Similarity Aggregation (GSA) for effective outlier filtration. The suggested technique is evaluated using the comprehensive CICIoT2023 dataset, comprising over 46 million samples and 34 unique attack types. Experimental results demonstrate outstanding performance, achieving 99.41% accuracy in assessments. This research introduces a scalable, efficient, and resilient federated learning approach tailored for actual IoT intrusion detection applications.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
R-IDF: Addressing the accuracy fallacy in evaluating LSTM-based intrusion detection
Phan Thanh HyDeep learning has become a cornerstone of Network Intrusion Detection Systems (NIDS), but its susceptibility to adversarial attacks reveals a major gap between reported accuracy and actual robustness. To close this gap, we introduce the Robustness-oriented Intrusion Detection Framework (R-IDF), which combines CTGAN-based data balancing, adversarial training strategies, diverse attack models, and a robustness metric suite. Evaluated on the CSE-CICIDS2018 dataset with LSTM-based models, R-IDF shows that although all models achieve nearly 99% clean accuracy, their retained accuracy under adaptive adversarial attacks often falls below 50%. This discrepancy, formalized as the Accuracy Fallacy Gap (AFG) and summarized by AUCRA, demonstrates that clean accuracy alone severely overestimates resilience. While adversarial training and defensive distillation improve resistance to transfer-based attacks, both fail against stronger adaptive adversaries. These results highlight the urgent need for robustness-oriented evaluation. By exposing hidden vulnerabilities and standardizing multi-attack assessment, R-IDF provides a reproducible benchmark that shifts NIDS evaluation beyond accuracy toward a more realistic measure of security.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Detecting "Nine-Dash Line" Images in Digital Content via Faster R-CNN and DINOv2-Based Knowledge Distillation
Do Tran TuThe "nine-dash line" image represents an illegal maritime claim that frequently appears in digital content, posing significant risks of distorting Vietnam's territorial sovereignty. Manual moderation cannot keep pace with the rapidly growing volume of data, especially as such infringing images are often small, thin, and subtly embedded. In this study, we employ an enhanced Faster R-CNN model, incorporating knowledge distillation (KD) from the vision-language model DINOv2 to detect "nine-dash line" imagery. A KD loss function is utilized to align embeddings between predicted and ground-truth regions, thereby improving the detection of small and occluded objects. The model is trained on a specialized dataset of 836 images and achieves a mean Average Precision (mAP) of 58.4\%, representing an improvement of 6.2\% over the baseline Faster R-CNN with a ResNet-50 backbone. In addition, the employed approach achieves an Average Recall (AR) of 66.2\%, compared to 58.8\% for the baseline model. This system demonstrates strong potential for integration into automated digital content moderation workflows, supporting the protection of national sovereignty in cyberspace.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
A Smart Curriculum Vitae Analysis and Recommendation System for Job Application Support
Lai Quang VinhIn the context of a rapidly evolving digital economy, the recruitment process is increasingly relying on automated systems and data-driven decision-making. The Curriculum Vitae (CV) remains a central element in evaluating candidates; yet, many job seekers, especially students and recent graduates, face challenges in aligning their profiles with the expectations of employers. Existing CV creation tools mainly emphasize design aesthetics and template formatting, while lacking semantic analysis, personalized feedback, and skill development pathways. This study introduces a smart CV analysis and recommendation system that integrates techniques from Natural Language Processing (NLP), Optical Character Recognition (OCR), and advanced large-scale language models (LLMs). The system is capable of parsing and analyzing candidate CVs, measuring their relevance against job descriptions (JDs), identifying strengths and deficiencies, and generating customized skill development roadmaps accompanied by targeted practice exercises. Furthermore, it supports automated CV generation in standardized formats optimized for applicant tracking systems. Experimental evaluation with a cohort of final-year university students demonstrates that the system substantially improves the alignment between CVs and job requirements, reduces preparation time, and enhances candidate confidence. These findings highlight the potential of AI-driven CV analysis systems as a bridge between academic training and labor market demands.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Deep Learning-Based Recognition and Classification of Technical Errors in Squat Movements.
Le Mau Hai DangThis study presents a simple and effective approach based on the application of computer vision combined with machine learning and deep learning models to identify and detect common errors in performing squat exercises. The model was trained on a dataset composed of self-recorded and online videos, video segmented into complete workout sequences, and categorized into four main error groups. Our method uses pose estimation to extract skeletal keypoints and compute joint angles such as knee angle, torso angle, and back angle, which are then processed by a Gated Recurrent Unit (GRU) model integrated with an attention mechanism to recognize and classify technical errors. The experimental results demonstrate that the model can detect errors with high accuracy, helping gym practitioners improve their technique and reduce the risk of injury during training.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
OWLViz: An Open-World Benchmark for Visual Question Answering
Thuy NguyenWe present OWLViz, a challenging benchmark for Open WorLd VISual question answering that evaluates multimodal AI systems on realistic, practical tasks. OWLViz features 248 carefully curated questions requiring the integration of multiple capabilities: common-sense knowledge, visual understanding, web exploration, and specialized tool usage. The benchmark specifically challenges models with visually degraded inputs, complex multi-step reasoning involving counting and measurement operations, and knowledge-intensive queries requiring external information retrieval from minimal visual cues. While humans achieve 69.2% accuracy on these intuitive tasks in under one minute, even state-of-the-art VLMs struggle dramatically, with the best model, Gemini 2.5 Pro, achieving only 27.09% accuracy. Current tool-calling agents and GUI agents, which rely on vision and vision-language models as tools, perform even worse, often failing to engage with available tools effectively. This substantial performance gap reveals critical limitations in multimodal systems' ability to select appropriate tools, coordinate heterogeneous resources, and execute complex reasoning sequences. OWLViz establishes new directions for advancing practical, open-world AI research and agent development.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
SwahiliVQA: A Dataset for Visual Question Answering in Swahili Language
Mbwana Francis StephenThis paper introduces the first Swahili Visual Question Answering (SwahiliVQA) dataset, addressing the critical underrepresentation of African languages in vision-language research. Swahili, spoken by over 100 million people across East Africa, remains severely underserved in AI development despite its widespread use. Our dataset comprises 10,000 images paired with 41,448 question-answer combinations, translated from English resources and validated by native Swahili speakers to ensure linguistic authenticity. We establish baseline performance metrics using various model architectures and propose a novel multimodal approach that combines CLIP's vision encoder with a Swahili-specific RoBERTa model, achieving 38.38% accuracy. This work provides essential resources for Swahili-language AI development, establishes a methodological framework for creating VQA datasets in other underrepresented languages, and contributes to more equitable artificial intelligence that serves diverse linguistic communities across East Africa and beyond.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Bridging Context and Preference: A Hybrid RAG-SVD Framework for Personalized Conversational Recommendations
Đỗ Thị LiênPersonalized recommendations in dialogue systems require both contextual understanding and accurate modeling of user preferences. While Retrieval-Augmented Generation (RAG) excels at generating context-aware responses by retrieving semantically similar conversations, it often lacks personaliza-tion grounded in user history. Conversely, collaborative filtering methods such as Singular Value Decomposition (SVD) effectively capture long-term user preferences but are limited in handling dynamic conversational context. In this paper, we propose a hybrid framework that bridges contextual rele-vance and preference modeling by integrating RAG with SVD for personal-ized conversational recommendations. Our system retrieves relevant dialogue snippets using FAISS and combines them with item suggestions generated from a user–item interaction matrix via SVD. These signals are fused into a prompt for a generative language model, enabling them to produce fluent, context-sensitive, and personalized recommendations, even under cold-start conditions where user–item interactions are extremely sparse. Empirical evaluations on the ReDial dataset demonstrate that our framework signifi-cantly improves recommendation accuracy compared to baseline RAG-only, Generative-only, SVD-only systems, confirming the effectiveness of com-bining retrieval-based context modeling with collaborative filtering in con-versational recommendation tasks.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
A Hybrid Machine Learning and XAI Architecture for Intelligent Career Guidance Systems
Lê Mạnh HàIndustry 4.0 and artificial intelligence are shown to bring great change or disappear a large proportion of work, while new jobs are born With the dynamical background comes the demand for a smart career guidance system, which will provide advice that is personalized, reliable, and adaptive. This paper systematically reviews the literature on the application of machine learning (ML) and explainable artificial intelligence (XAI) to career guidance. On the basis of PRISMA guidelines, 847 documents published between 2019 and 2025 were carefully screened, and 95 high-quality articles were extracted for in-depth review. The review classifies main ML methods—including collaborative filtering, content-based filtering, deep learning architectures such as LSTM, Transformer, and GNN, reinforcement learning, and their performance, limitations, and interpretability were judged. In parallel, the author analyzes such essential but questioned XAI techniques as LIME, SHAP, attention mechanisms, decision rules and counterfactual explanations in terms of their transparency and perceived user trust, as well as how easily acted upon these explanations are. From these foundations, the paper presents a five-layer hybrid ML-XAI framework that integrates data processing, knowledge maps, ensemble ML models, multi-level explanations, and user-centered presentation. In addition to these, future developments, such as flat or formidable language models and federated learning for maintaining privacy and fairness-aware algorithms, are explored, together with key challenges for further research. All in all, the paper provides a structured basis and practical guidance for next-generation, intelligent, transparent, and equitable career guidance systems.
Năm:2025
|
Chủ đề: Kỹ thuật điện tử và viễn thông
|
Nhà xuất bản: International Journal of Science, Engineering and Technology
The Mechanism of Innovation Ecosystem's Influence on E-commerce Competitiveness in Vietnam: An Analysis of Four Key Pillars
Nguyễn Thị HồngThis study explores the role of the innovation ecosystem in enhancing the competitiveness of e-commerce enterprises in Vietnam. Based on secondary data analysis, the paper synthesizes insights from reputable sources such as the Vietnam E-Business Index Report, e- Conomy SEA, the Vietnam Innovation and Startup Report, as well as international academic literature. The findings indicate that the competitiveness of e-commerce enterprises is not solely determined by internal resources but also depends on the level of development and coordination among four components of the innovation ecosystem: technological infrastructure, complementary resources, organizational connectivity, and institutional policies. Digital infrastructure contributes to operational optimization; complementary resources such as logistics, payment, and digital marketing enhance value chain flexibility; connections with incubators, venture capital funds, and research - academic institutions foster
innovation capacity; and institutional policies shape a transparent and sustainable competitive environment. This paper contributes to the theoretical discourse by clarifying the mechanisms linking the innovation ecosystem and enterprise competitiveness, while also offering
managerial implications. E-commerce firms are advised to proactively leverage ecosystem resources, whereas the government should continue to improve the policy framework and digital infrastructure, thereby creating a fair competitive environment and promoting sustainable development.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Unshared Weight combine with Self-attention at Base model for On-The-Fly Fine-Grained SBIR
Nguyễn Hữu TuấnOver the years, on-the-fly fine-grained sketch-based image retrieval (SBIR) has demonstrated significant potential for real-world applications. However, previous methods rely on weight-sharing and attention mechanisms, which increase the computational load and reduce accuracy. This study enhances feature extraction and task differentiation by eliminating weight sharing, thereby preserving strong feature discriminative power. Our model achieves 82.04% Acc@5 and 92.57% Acc@10 on QMUL-Chair-V2, and 66.37% Acc@5 and 80.78% Acc@10 on QMUL-Shoe-V2, outperforming Bi-LSTM, Multi-granularity and LGRL models, demonstrating improved retrieval performance.
Năm:2025
|
Chủ đề: Khoa học kỹ thuật và công nghệ khác
|
Nhà xuất bản: Signal, Image and Video Processing
Detection of Dental Decay in children using Modern Deep Learning Models
Nguyễn Hữu TuấnDental decay is one of the most common oral health problems worldwide, but datasets about it are very limited. This paper introduced a new dataset for dental decay, which included images of children over 12 years old from many dental facilities in Viet Nam. At the same time, we also introduced a new change learning rate method for SGD optimizer. To verify the effectiveness of the proposed method, we compare the results with the Adam optimizer to evaluate the differences in convergence speed and accuracy based on training 3 models: Faster R-CNN, YOLOv3 and DETR, without using pretrained weights
Năm:2025
|
Chủ đề: Khoa học kỹ thuật và công nghệ khác
|
Nhà xuất bản: Journal of Information Hiding and Multimedia Signal Processing
Determinants of Digital Banking Adoption Intentions: Evidence from Vietnamese commercial banks
Trần An QuânSince the 1990s, digital banking in Vietnam has grown rapidly due to widespread smartphones
use and internet connectivity, offering convenience and efficiency. However, cash usage and
preference for traditional banking persist as barriers. This study examines key determinants
influencing customers’ intentions to adopt digital banking services in seven Vietnamese
commercial banks to enhance digital transformation strategies. Primary data were gathered using
a standardized questionnaire from 322 customers across seven commercial banks in Vietnam,
analyzed using linear regression. The findings indicate that perceived trust, usefulness, ease of
use, and convenience significantly encourage customers’ intentions to adopt digital banking
services, with trust exerting the most substantial influence. Conversely, perceived risk negatively
affects adoption intentions and other related factors. Although internal validity is supported by
rigorous statistical analysis, the external validity is constrained by the non-probability sampling
method, meaning findings are indicative for similar large commercial banks rather than the
Vietnamese banking system overall. Despite these limitations, this study provides valuable and
insightful contributions, enabling these major banks to refine their digital transformation
strategies, enhance customer experiences, and respond more effectively to diverse customer needs
in the rapidly evolving digital banking landscape.
Năm:2025
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản:
Application of Machine Learning Algorithms for car type classification in Vietnam Market
Trần An QuânSupervised learning is one of the big problems of machine learning that has been applied in many fields. This paper applies it for classification problem on car type in Vietnam market. It uses the database of car industry in Vietnam in 2023, refers to car specification to make prediction of car types. The paper applies 6 algorithms, including: Logistic Regression, Naive Bayes, Linear Discriminant Analysis, Classification and Regression Tree, Support Vector Machine and Gradient Boosting for model training and testing. After selecting the most appropriate algorithm, the research evaluates efficiency of classification in machine learning by accuracy score and confusion matrix.
Năm:2025
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản: IJARIIE
Two-stage APT malware propagation model in computer networks
Đỗ Xuân ChợEarly detection and prevention of advanced persistent threats (APT) is a critical challenge in cybersecurity. This paper presents an innovative approach using dual susceptible–infected–recovered (Dual-SIR) model to predict the two-stage spread of APT malware within networks. The first SIR model addresses infections at the first stage—device and user level, serving as a precursor to server compromise. The second SIR model focuses on the second stage of propagation—server infections, where sensitive organizational data is stored. Experimental results demonstrate the effectiveness of our proposed model not only for APT malware but also for other types of malware. Our work significantly contributes to the field of cybersecurity by offering a more accurate and proactive method for predicting malware spread. Additionally, this approach has potential applications in forecasting the dissemination of malware in wireless sensor networks and the spread of malicious information on social media platforms.
Năm:2025
|
Chủ đề: Khoa học máy tính và thông tin
|
Nhà xuất bản: Neural Computing and Applications
A novel ensemble learning-based model for APT attack detection
Vũ Thành ĐứcImproving the effectiveness of APT attack detection models is one of the most critical and essential tasks today. Following this trend, this paper proposes a new model called ACDF-mLSTM to address two primary challenges currently faced by research in this field: (i) data imbalance and (ii) information aggregation and feature extraction. Specifically, to solve the data imbalance problem, the paper proposes a novel data generation method named ACDF. This method leverages advanced and sophisticated techniques to focus on identifying crucial points in sequential data and analyzing context by considering preceding and succeeding data points. Subsequently, a Diffusion Model is applied to generate synthetic APT attack data, built upon the principle of gradual diffusion. With this approach, the ACDF model can generate more meaningful and realistic data. Next, to address the task of information aggregation and feature extraction, the paper proposes a new deep learning model named mLSTM, based on the optimization of Long Short-Term Memory (LSTM). Thus, the mLSTM model performs two main tasks: (i) extracting information from network flows within traffic and (ii) aggregating and highlighting important information before it enters the classification model. In the experimental section, the paper evaluates the ACDF-mLSTM model for the first time across various scenarios and datasets to demonstrate its effectiveness and adaptability. The evaluation results show that the ACDF-mLSTM model outperformed most other methods by an average of 2 to 12% across all metrics and on all experimental datasets.
Năm:2025
|
Chủ đề: Khoa học máy tính và thông tin
|
Nhà xuất bản: Memetic Computing
A Study of Multilingual Approaches to Vietnamese Information Retrieval
Nguyễn Thị Tuyết HảiNatural language processing (NLP) applications have an extensive development in various languages. However, information retrieval (IR) for Vietnamese texts has not yet received adequate attention from the research community. In this study, we fne-tune high-quality multilingual embedding models to enhance the effectiveness of document retrieval. Additionally, we investigate several approaches to information retrieval in order to gain deeper insights into their performance and applicability.
Năm:
|
Chủ đề: Trí tuệ nhân tạo
|
Nhà xuất bản: