Journal of Web Engineering

Hybrid Layered Retrieval and Task-aware Embeddings for Efficient Persian RAG Systems

2025-11-07T06:50:57+01:00

This study introduces task-injected layered hybrid retrieval-augmented generation (TILHR-RAG), a framework specifically designed for Persian to address the scarcity of native-language resources and the limitations of English-centric approaches. The architecture combines task-aware query augmentation, a layered retrieval strategy, and a hybrid semantic–lexical retriever, all supported by a multi-stage pipeline that includes preprocessing, document chunking, question generation, and embedding. A novel mechanism for injecting task-specific vectors directs retrieval toward domain intent while preserving comparability across documents. The layered design operates in three stages: per-task frequently asked questions (FAQ) retrieval, hybrid document search using FAISS semantic similarity combined with BM25 keyword matching, and a fallback response generated by a large language model (LLM). This structure ensures both precision and robustness. Comprehensive experiments across five progressively refined configurations demonstrate that TILHR-RAG achieves the best balance among accuracy, efficiency, and scalability, reaching 89.67% semantic accuracy with moderate latency and memory consumption on NVIDIA A100 hardware. Further evaluations on low-resource graphics processing units (GPUs) confirm that accuracy remains stable under hardware constraints, although latency increases significantly. Moreover, multilingual E5 embedding models substantially improve retrieval and generation quality for Persian – outperforming ParsBERT and Sentence-BERT (SBERT) – by mitigating challenges such as orthographic variation and complex compound word structures. Taken together, these findings establish task-injected layered hybrid retrieval-augmented generation as a practical, reproducible, and resource-efficient blueprint for Persian question answering, advancing retrieval-augmented generation for low-resource languages without requiring costly large language model fine-tuning, while also offering adaptable strategies for broader multilingual applications.

The Case for HTML First Web Development

2025-11-30T17:06:19+01:00

Since its introduction, the web has become the largest application platform available globally. HyperText Markup Language (HTML) has been an essential part of the web since the beginning, as it allows defining webpages in a tree-like manner, including semantics and content. Although the web was never meant to be an application platform, it evolved as such, especially since the early 2000s, as web application frameworks became available. While the emergence of frameworks made it easier than ever to develop complex applications, it also put HTML on the back burner. As web standards caught up, especially with milestones such as HTML5, the gap between the web platform and frameworks was reduced. HTML First development emphasizes this shift and puts focus on literally using HTML first when possible, while encouraging minimalism familiar from the early days of the web. It seems HTML-oriented web development can provide clear benefits to developers, especially when it is combined with complementary approaches, such as embracing hypermedia and moving a large part of application logic to the server side. In the context of the htmx project, it was observed that moving towards HTML can reduce the size of a codebase greatly while leading to maintenance and development benefits due to the increased conceptual simplicity. Holotype-based comparisons for content-oriented websites show performance benefits, and the same observation was confirmed by a small case study where the Yle website was converted to follow HTML First principles. In short, the HTML First approach seems to have clear advantages for web developers, while there are open questions related to the magnitude of the benefits and the alignment with the recent trend of AI-driven web development.

Quantum Random Number Generators for NIST Post-Quantum Cryptography Standard Algorithms

2026-01-15T11:08:18+01:00

In recent years, the advancement of quantum computing technology has posed potential security threats to RSA cryptography and elliptic curve cryptography. In response, the National Institute of Standards and Technology (NIST) published several Federal Information Processing Standards (FIPS) of post-quantum cryptography (PQC) in August 2024, including the Module-Lattice-Based Key-Encapsulation Mechanism (ML-KEM), Module-Lattice-Based Digital Signature Algorithm (ML-DSA), and Stateless Hash-Based Digital Signature Algorithm (SLH-DSA). Although these PQC algorithms are designed to resist quantum computing attacks, they may not provide adequate security in certain specialized application scenarios. To address this issue, this study proposes quantum random number generator (QRNG)-based PQC algorithms. These algorithms leverage quantum computing to generate random numbers, which serve as the foundation for key pair generation, key encapsulation, and digital signature generation. A generalized architecture of QRNG is proposed, along with the design of six QRNGs. Each generator is evaluated according to the statistical validation procedures outlined in NIST SP 800-90B, including tests for verification of entropy sources and independent and identically distributed (IID) outputs. Experimental results assess the computation time of the six QRNGs, as well as the performance of QRNG-based ML-KEM, QRNG-based ML-DSA, and QRNG-based SLH-DSA. These findings provide valuable reference data for future deployment of PQC-based Transport Layer Security in web systems.

Application of ZKML for Unpredictive Epidemic Response

2025-05-21T09:21:29+02:00

We build and evaluate a concrete Zero-Knowledge Machine Learning (ZKML)-based pipeline for epidemic diagnosis and show that it can enforce computational integrity without exposing raw medical data in a Web3 setting. In response to security challenges posed by centralized data handling in medical AI applications, particularly during public health crises such as COVID-19, ZKML offers a privacy-preserving alternative by combining machine learning and Zero-Knowledge Proofs (ZKP). We experimentally applied ZKML to a CNN (Convolutional Neural Networks)-based COVID-19 diagnostic model, achieving 87% accuracy and 0.35 loss. All proof generation and verification processes were executed entirely off-chain, with the verified outputs represented as committed public_vals recorded on-chain via smart contracts. To ensure authenticity, the system enforces dual ECDSA signature verification from both the model provider and the data provider. This mechanism prevents unauthorized submissions and confirms the validity of the result before it is stored on-chain. The system was tested under both normal and adversarial conditions, demonstrating robust and reliable operation. By enabling decentralized trust and self-sovereign control over data, this architecture aligns well with Web3 principles. The results indicate that ZKML can support the development of privacy-preserving and verifiable AI systems.

ChainMark: Integrating an Invertible Neural Network and Blockchain for Ensuring Ownership Rights in Image Watermarking

2026-01-07T20:21:40+01:00

The widespread use of generative AI has intensified issues related to digital ownership, even as it enhances the efficiency of digital content creation. While watermarking is a primary method for protecting these rights, existing neural network-based approaches often prioritize robustness and imperceptibility, neglecting verifiable ownership. To address this limitation, this paper proposes ChainMark, a system that integrates an invertible neural network (INN) with blockchain technology. ChainMark employs an INN trained within the discrete wavelet transform domain to embed watermarks that are resilient to diverse signal processing attacks. Crucially, unlike traditional approaches that rely solely on watermark extraction, the proposed system secures the verification process through a blockchain smart contract. Experimental results validate the system’s theoretical security and demonstrate that the joint LH-HL model configuration achieves an optimal trade-off between visual quality and extraction accuracy. Consequently, ChainMark effectively guarantees creator rights by ensuring both high-performance watermarking and trustworthy ownership verification.

Blockchain-Assisted Privacy-Preserving Retrieval for Sensitive Outsourced Data

2026-04-13T09:25:19+02:00

With the continuous outsourcing of sensitive data to cloud platforms and third-party environments, how to achieve efficient and trustworthy data retrieval while preserving data confidentiality and query privacy has become a critical issue in outsourced data security management. Existing studies mainly focus on secure storage and privacy-preserving retrieval but suffer from limitations in the efficiency of multi-attribute conjunctive queries, the suppression of intermediate information leakage during auxiliary condition verification, and the trustworthy traceability of the retrieval process. To address these issues, this paper proposes a blockchain-assisted privacy-preserving retrieval framework for sensitive outsourced data. The framework adopts a collaborative paradigm of off-chain storage and retrieval together with on-chain commitment and auditing. By combining a frequency-aware primary search term selection strategy with a concealed auxiliary verification mechanism, it enables secure filtering of task-relevant data under multi-attribute conditions. Meanwhile, by recording index commitments, query digests, and result digests on the blockchain, the proposed framework enhances the verifiability and traceability of index states and access processes. Security analysis and experimental results demonstrate that the proposed scheme can achieve more efficient construction of auxiliary decision structures, more stable query performance, and better storage overhead while preserving data confidentiality and query privacy.

Combination Optimization of Web Services with Credibility and QoS Attributes

2025-11-26T21:06:14+01:00

To address the problem of low service quality and efficiency caused by non-functional attributes in traditional Web service composition, this paper proposes a Web service composition model construction and optimization method that integrates trustworthiness calculation and QoS attributes. First, we conduct separate objective reputation assessments and subjective trust evaluations based on user needs. Next, we perform attribute normalization and calculate subjective and objective weights according to QoS attribute classifications to establish a comprehensive weighting method. Finally, we develop a multi-attribute QoS composite service model and optimize it using the IBBS algorithm to select the optimal Web service combination. Experimental results demonstrate that both the proposed reliability calculation method and QoS attribute service selection approach can effectively compute Web service reliability and optimize candidate service selections. Compared with advanced optimization algorithms, the IBBS algorithm exhibits significantly better time efficiency and optimization outcomes. When implemented in the constructed Web service combination system, this algorithm substantially enhances service quality, better meets user requirements, and demonstrates robust effectiveness and stability.

A Semantic-Web-Driven Visualization and Fault-Reasoning Framework for Circuit-Oriented Systems

2026-01-08T20:07:58+01:00

Efficient and explainable reasoning over dynamic, heterogeneous data remains a key challenge for intelligent diagnostic and monitoring systems. This paper presents a unified semantic-web-based framework that integrates ontology modeling, rule-driven inference, and interactive visualization into a scalable, service-oriented architecture. The proposed system couples Web Ontology Language (OWL)-based knowledge representation with dynamic Semantic Web Rule Language (SWRL) rule execution and control-theoretic feedback, forming a closed-loop semantic reasoning cycle that continuously refines ontology and rule parameters. To ensure real-time performance, the framework employs parallelized rule evaluation, adaptive caching, and incremental inference across distributed reasoning nodes. A modular semantic query interface bridges reasoning and visualization layers, enabling transparent inspection of causal relationships and human-in-the-loop knowledge refinement. Experimental results demonstrate that the proposed system achieves sub-linear latency growth with ontology size, reduces inference delay by up to 56% through indexing-caching synergy, and maintains detection accuracy above 95% under complex fault conditions. The end-to-end latency remains below 300 ms for medium-scale ontologies, validating its suitability for real-time diagnostic, telepresence, and edge-analytics applications. These findings establish a novel synthesis between symbolic reasoning and adaptive system control, offering both computational efficiency and semantic interpretability for circuit-oriented and cyber-physical diagnostic systems, while providing a foundation that may be extended to other semantic reasoning domains.

Artificial Intelligence-Based Anomaly Detection for Large-Scale Web Data Security Monitoring

2026-02-24T03:55:06+01:00

With the rapid development of the World Wide Web and the popularity of Internet applications, the generation and exchange of data have exploded. Large-scale data generation and transmission also bring severe security challenges. In response to the problems that existing anomaly detection methods are difficult to jointly model the semantic context and temporal dependencies in non-encrypted scenarios, and that single-modal feature information is insufficient in encrypted scenarios, resulting in limited detection accuracy, this study proposes two artificial intelligence anomaly detection methods that are adapted to different scenarios. For non-encrypted/low-encrypted scenarios, a BERT-LSTM-TextCNN parallel fusion architecture is proposed. This architecture extracts high-order semantic features, long-term dependency features, and multi-scale local features through parallel branches, and achieves complementary enhancement of multi-perspective information through feature concatenation, effectively solving the problem of difficult collaborative modeling of multiple types of features in non-encrypted scenarios. For multi-encrypted scenarios, a detection method based on improved ResNet and cross-modal feature fusion is proposed. Different from traditional methods that only rely on deep learning features, the study adaptively weights and fuses the deep semantic features extracted by ResNet with flow statistics features and temporal features and optimizes the fusion weights through a learnable random forest, breaking through the bottleneck of insufficient single-modal feature information in encrypted traffic. The precision reached 97.18%, the recall rate reached 95.26%, and the F1-score reached 96.21%. The AUC values were all greater than 0.97, the false positive rate was 8.12% lower than the traditional method, and the single-batch data detection time was only 37.25 s. In the multi-encryption scenario, the precision, recall rate and F1-score of the cross-modal feature fusion method were 98.48%, 87.30% and 92.57%, respectively. This effectively solves the detection limitations caused by feature ambiguity in encrypted environments. In summary, the artificial intelligence anomaly detection method effectively improves detection accuracy and efficiency and provides a feasible technical path for building a comprehensive World Wide Web data security monitoring system.

Evaluation of the Usability of a Web Application for AI-Enhanced Multilingual Learning Platform: Based on the Indicator System of Language Cognitive Load Learning Efficiency

2026-03-04T17:14:06+01:00

This paper focuses on the many problems that exist in the evaluation of AI to improve the usability of Web applications for multilingual learning platforms, including the poor alignment effect between learning efficiency and usability, the imbalance of cognitive load regulation, and the singularity of evaluation indicators. In order to solve these problems, this study innovatively constructs a multi-dimensional evaluation index system integrating language cognitive load and learning efficiency and designs a dynamic evaluation model AILA-WA driven by AI. This model can combine learning algorithms to interact with data from Web applications and can collect real-time data related to language learning behavior and cognitive state feedback data from Web applications. It enables the optimization direction of Web applications to be identified and accurately quantified. Subsequent experiments successfully prove that the index system and the evaluation model can effectively improve the comprehensiveness accuracy of the evaluation of Web application usability. For example, in the scenario of multilingual learning, the cognitive load fitting deviation rate of the Web application using the AILA-model is the best compared to the Web application using the comparative model. At the same time, learning efficiency and CSAT user satisfaction are also at the level; and the model is suitable for Web applications. System response delay on multiple terminals is reduced to 0.3 s. These breakthroughs provide strong support for design iteration and usability optimization of AI to improve multilingual learning Web applications.