Journal of Web Engineering

IaaS Cloud Adaptive Anomaly Detection Based on the DQN Algorithm

Li Chen — 2026-05-24

Aiming at the challenges of anomaly detection of virtual machine memory, network, CPU and hard disk in the IaaS cloud environment, this study proposes an adaptive anomaly detection system based on a deep Q-network. The system constructs a hierarchical detection framework: a spatio-temporal feature extraction module via fused temporal convolutional networks (TCN) for sequential pattern mining and convolutional neural networks (CNN) for cross-metric correlation learning; a transfer learning module to enhance generalization; and a deep Q-network (DQN) based central controller that dynamically adjusts detection parameters through reinforcement learning. This architecture integrates with cloud workload schedulers by operating at the VM-level (anomaly detection) and edge-server level (DQN control), minimizing core network overhead. Experiments show that the research method achieves a detection accuracy rate of 99.8% in the benchmark test, with an F1 score of 98.7%, which is significantly superior to the accuracy rate of 96.5% of the single convolutional neural network, 92.3% of the multi-layer perceptron, and 97.8% of Google Net. The transfer training experiments show that the accuracy rate of the untuned model on the new dataset is only 70% to 80%, while the detection accuracy can be stably improved to 98% through the adaptive system driven by the DQN. The system shows low volatility during the dynamic adjustment process. The number of training iterations is reduced by 32.3% to 69.8% compared with the traditional static model, indicating that the research method does not affect the time complexity. Research shows that this framework effectively solves the problem of insufficient adaptability of static models to unknown data in the cloud environment through the collaborative mechanism of spatio-temporal feature extraction and reinforcement learning decision-making, providing intelligent operation and maintenance solutions for fields with high reliability requirements such as finance and healthcare.

HBCMS: A Web-native Hierarchical Blockchain CMS for OTT – Contract-aware APIs, Edge-cache Consistency, and HTTP-level SLOs

Suhwan Bae — 2026-05-24

Over-the-top (OTT) platforms must expose metadata and digital rights from numerous content providers (CPs) through the web while maintaining low latency and verifiable integrity. This paper presents HBCMS, a hierarchical blockchain-based content management system that separates contract-aware governance in the upper layer from high-rate metadata management in CP-owned lower chains. The web layer is realized through a contract-aware API, a consistency model aligned with edge and browser caching, and HTTP-level service-level objectives (SLOs) linking blockchain verification to observable web behavior. The upper and lower chains are connected via Merkle-root anchoring, and verification proceeds through contract validation, anchor matching, and Merkle proof verification exposed as RESTful endpoints. Anchoring is modeled as a Poisson process that determines the rate required to satisfy verification windows and guides content delivery network (CDN) cache-control policies. In large-scale experiments with up to 1000 CPs, HBCMS achieved about 2.6 k transactions per second (TPS), 0.185 s end-to-end latency, and 99.4% verification success, with lower-chain queries dominating delay. These results provide reproducible guidance for API versioning, cache invalidation, and observability in scalable OTT web architectures.

Research on the Propagation and Topic Mining of Online Public Opinion in Social Networks

Gaoyue Rong — 2026-05-24

Online public opinion has become a critical component of web-based social systems, where large-scale user interactions generate complex propagation behaviors and evolving topic structures. With the rapid growth of social networking platforms, public opinion exhibits network-driven diffusion, temporal volatility, and fragmented topic evolution, posing challenges for web platform monitoring and governance. Existing studies typically rely on either epidemic propagation models or standalone topic modeling methods, limiting their ability to jointly capture diffusion mechanisms and content evolution. To address this issue, this study proposes an integrated web analytics framework that combines epidemic-based propagation modeling with topic mining. Using real data from the Weibo platform, an improved epidemic dynamics model is developed to simulate opinion diffusion over complex networks, with parameters calibrated from observed user interactions. In parallel, latent Dirichlet allocation (LDA) is applied to large-scale textual data to extract latent topics and analyze their temporal evolution. The results show that the network positions of initial propagators and key topological characteristics significantly influence propagation dynamics. Topic mining further reveals six stable thematic clusters with distinct evolutionary patterns across time windows. The proposed framework provides an interpretable system-level approach for analyzing online public opinion, offering practical support for real-time monitoring, moderation workflows, and decision-support systems in web governance.

Metaheuristics for Ontology-Based Information Extraction Rule Learning

Michel Capelle — 2026-05-24

The Semantic Web aims to make information intelligible for computers. In the Semantic Web, unstructured information from text is represented using ontologies, such that computers can understand text better. However, adding text information to existing ontologies by hand is time-consuming. Information extraction rules can help to automate this process. In the process of learning information extraction rules, patterns are constructed that consist of lexico-syntactic and lexico-semantic features from text, which aim to extract Resource Description Framework subject-predicate-object expressions. In this paper, we investigate the following four metaheuristics for learning ontology-based information extraction rules: Particle Swarm Optimization, 2-Phase Optimization, Ant Colony Optimization, and Genetic Algorithm (GA). We evaluate all methods using financial news data. GA gives the best F₁-measure results, but the other metaheuristics are faster.

A Service-integrated Web Framework Supporting Reliability Tracing in Smart Distribution Networks

Haiyan Wang — 2026-05-24

Smart distribution networks (SDNs) now integrate more distributed energy resources, IoT devices, and multi-stakeholder systems, raising service collaboration complexity. This creates key challenges for real-time fault localization, cross-organizational service compatibility, and performance oversight. This paper presents a novel service-integrated web framework designed to address the challenges of reliability tracing, service integration, and performance monitoring SDNs. The framework leverages a modular architecture that integrates advanced methodologies, including a service-oriented architecture (SOA) for seamless cross-organizational collaboration, metadata management using semantic web technologies for enhanced interoperability, and real-time performance monitoring with anomaly detection. An end-to-end reliability tracing mechanism that combines event logging with causal relationship analysis is implemented to localize faults with high accuracy. The development process adopts a model-driven approach, utilizing UML and SysML for architectural modeling, and employs containerized deployment via Kubernetes for scalability. Unlike existing web-based reliability management systems that operate as isolated analytics or visualization layers, the proposed framework integrates service orchestration, semantic metadata reasoning, and fault-tracing analytics into a unified architecture. This service-integrated design enables end-to-end information flow – from data acquisition to reliability inference – under a common web infrastructure, representing a substantive advancement in the web engineering of power system reliability applications. Experimental validation in a simulated SDN environment demonstrates that the framework achieves a reliability tracing accuracy of 97.2%, a detection-to-reporting time of 1.8 s, and resource utilization increases of less than 5% per node. These metrics – tracing accuracy, latency, and resource efficiency – are directly aligned with the reliability evaluation indices defined in IEEE 762 and IEC 62559 standards for smart distribution networks, ensuring comparability with established system reliability benchmarks. These results highlight the framework’s ability to meet the demands of dynamic distributed systems while providing a foundation for future advancements.

Design and Implementation of a High-availability Network Infrastructure for Large-scale Interactive E-learning

Xiaoyong Su — 2026-05-24

Large-scale interactive online education platforms present significant challenges to traditional elastic scaling strategies based on static thresholds due to their dynamic and unpredictable load characteristics. This article designs and implements a cloud native high-availability network infrastructure centered around an intelligent elastic scaling model that integrates time series prediction and reinforcement learning. This architecture deeply integrates microservices and service mesh technology, predicting short-term resource requirements through historical load and contextual information (such as course schedules), and driving Kubernetes clusters to perform pre-scaling. The research is validated through simulation analysis and real prototype system experiments. The results show that in the simulation environment, the model improves resource prediction accuracy by 25% compared to traditional Horizontal Pod Autoscaler (HPA) strategies, and reduces service level agreement (SLA) violation rates by more than 60% during sudden traffic. In practical systems, the average response delay during peak periods is reduced by 40%, resource utilization increases by 35%, and system availability reaches 99.99%, significantly improving service quality and resource utilization efficiency.

Knowledge Graph-augmented Sequential Recommendation with Adaptive Time-decay Kernels

Xuelian Zhang — 2026-05-24

To address the limitations of existing knowledge graph-enhanced recommendation systems – particularly their reliance on static fusion mechanisms that fail to capture the dynamic evolution of user interests and their inadequate modeling of heterogeneous information interactions – this paper proposes AdaTKGR, an adaptive time-decay weighted framework for knowledge graph-enhanced recommendations. First, a time-aware self-attention mechanism is introduced to effectively model temporal dependencies in user behavior sequences, thereby capturing fine-grained patterns of interest shift over time. Second, we integrate the RippleNet-style knowledge propagation strategy with a learnable temporal decay kernel, enabling dual-weighted representation learning based on both relational distance within the knowledge graph and temporal recency. Third, a cross-compression unit leveraging low-rank bilinear transformations is designed to facilitate deep semantic interaction between user–item interaction embeddings and knowledge graph entity representations. Finally, a time-gated multi-task learning objective is formulated to dynamically balance the primary recommendation task with auxiliary knowledge graph link prediction, enhancing joint optimization. Extensive experiments are conducted on three benchmark datasets – Book-Crossing, Last-FM, and MovieLens-1M – where AdaTKGR achieves average improvements of 6.1% and 8.4% in HR@10 and NDCG@10, respectively, over the strongest baseline methods. Notably, the proposed framework exhibits enhanced generalization performance and interpretability, particularly under data-sparse conditions. This work presents a principled approach to jointly optimizing temporal dynamics modeling and semantic knowledge integration in recommender systems.

Lightweight Probabilistic RL for Web-app Compatible Large-scale OHT Path Optimization

OkHwan Bae — 2026-05-24

As modern smart-factory environments increasingly require real-time remote operation and lightweight cloud-based control, routing intelligence for OHT systems must be fully web-app compatible, supporting scalable deployment without reliance on high-end local infrastructure. To address these demands and the limitations of static algorithms in large-scale OHT systems, this study proposes a multi-agent reinforcement learning model based on proximal policy optimization, incorporating a state space that accounts for chain blockage probability. The key metric, “movement success probability,” integrates preceding agent states to predictively assess chain-reaction congestion, enabling agents to proactively select stable detours. To enhance scalability in high-density environments, the model stabilizes learning through a lightweight policy initialization approach rather than requiring large-scale training from scratch. Moreover, the proposed decentralized structure minimizes central computational overhead, aligning naturally with web-app deployment and enabling real-time monitoring across distributed environments.

In a simulation with 1333 nodes and 100 OHTs, the proposed model achieved an average task completion distance of 166,809 mm, improving efficiency by 4.1% over the rule-based Floyd–Warshall method (173,940 mm). Notably, in worst-case scenarios where the rule-based method surged to 321,753 mm due to congestion, the AI model maintained 176,268 mm, achieving a 45.2% reduction and demonstrating superior operational stability.

Latent Diffusion Models: A Survey on Foundations, Variants, and Web-scale Deployments

Jee-Woo Shin — 2026-05-24

Latent diffusion models (LDMs) have rapidly become the de facto backbone of web-scale generative systems, powering text-to-image platforms such as Stable Diffusion and their video, 3D, and domain-specific extensions. By performing the diffusion process in a compressed latent space rather than directly in pixel space, LDMs achieve a favorable trade-off between computational efficiency and generative fidelity, enabling deployment in interactive web applications and large-scale content pipelines. This paper presents a comprehensive survey of LDMs from the perspective of both foundational modeling and web engineering. We first review the background of diffusion models and latent representations, contrasting LDMs with classical VAEs, GANs, and pixel-space diffusion models. We then dissect the architectural design of LDMs, including autoencoder backbones, latent-space U-Nets and diffusion transformers, conditioning mechanisms, training objectives, and sampling accelerations. Building on recent general surveys of diffusion models in vision, temporal data, and inverse problems, we propose a taxonomy of LDM variants, covering 2D image models, video and 4D models, and domain-specific LDMs in medical imaging, watermarking, time series, and text. From a web engineering viewpoint, we analyze LDM-based services exposed via web APIs, hosted user interfaces, and developer platforms, and discuss system-level concerns such as scalability, latency, cost, safety, and governance. We review current evaluation methodologies (quality, diversity, downstream task performance, robustness, watermarking) and highlight open challenges in controllability, interpretability, resource efficiency, and regulatory compliance, especially in light of recent legal and societal developments around generative deepfakes and copyright. This survey aims to provide both a conceptual map of LDM research and practical guidance for designing, deploying, and governing LDM-driven web systems.