Libyan Journal of Informatics

A Comparative Study of Lexical, Dense, and Hybrid Retrieval for Bilingual Graduation Proposal Similarity Detection

Marwa Eddakhli — 2026-09-30

We compare three retrieval strategies for bilingual (Arabic--English) graduation project proposal similarity detection: a lexical baseline (BM25), a dense pipeline (fine-tuned Multilingual-E5 Bi-Encoder with Cross-Encoder re-ranking), and a hybrid pipeline that fuses both via Reciprocal Rank Fusion. The Bi-Encoder was domain-adapted on 6000 synthetic proposals and evaluated on an absolute gold standard of 300 real Arabic proposal pairs, translated to English via DeepL with human review, under a strict cross-lingual task where each English query must locate its exact Arabic counterpart among 300 candidates. BM25 exhibited severe performance degradation in the cross-lingual setting, achieving a Precision@1 of 0.113---effectively at the level of uninformed retrieval. Crucially, the Dense pipeline significantly outperformed the Hybrid pipeline (p < 0.001, Wilcoxon signed-rank test). In this strict cross-lingual environment, RRF promoted irrelevant BM25 keyword matches, causing the lexical component to introduce spurious candidates that degrade re-ranking precision by displacing correct dense candidates. Qualitative evaluation further revealed that Hybrid produces false positives on novel queries, whereas Dense correctly rejects them with negative Cross-Encoder scores. We recommend the pure Dense pipeline for institutional deployment due to its superior accuracy, false-positive robustness, and single-index simplicity. External validation on the STS17 benchmark confirms that the dense models possess genuine cross-lingual semantic understanding (0.77 on Arabic pairs).

Evaluating the Integration of AI and IoT in Mobile App Education

Khaled M. Ben Hamed — 2026-09-30

The integration of Artificial Intelligence (AI), Internet of Things (IoT), and Ambient Computing (AC) is transforming Mobile Application Development (MAD), requiring new pedagogical approaches for undergraduate IT programs. This paper presents a 14-week intermediate MAD course for Mobile Computing (MC), Software Engineering (SE), and Web Technologies (WT) students. The course adopts a hybrid design with innovative pedagogy, evaluated for its effectiveness in fostering student engagement and real-world competency. Of 52 enrolled students (26 MC, 26 WT), 46 completed all assessments (88.5%). Using a quasi-experimental post-test-only design, analysis of six assessment types (mean = 62.89, SD = 18.78) found no significant program effect on pass/fail (χ²(1, N = 46) = 0.26, p = .610) or gender effect (χ²(1, N = 46) = 0.48, p = .489). Survey results revealed high satisfaction (94%), teamwork dominance (89%), AI adoption barriers (36%), and confidence disparities. Findings indicate the hybrid design accommodated heterogeneous preparation, with coursework predicting final exam performance (β = .57, p < .001, R² = .40) and early assignment performance serving as an early-warning indicator.

IPv4 vs IPv6 Efficiency Evaluation of Triple Redistribution

Hanan Dagez — 2026-09-30

This study aims to evaluate the impact of incorporating the IS-IS protocol into the dual redistribution chain (EIGRP/OSPF) to create a triple redistribution sequence. It also investigates the efficacy of this scenario in IPv4 and IPv6 settings inside IP/MPLS core networks. Maintaining reliable Quality of Service (QoS) for voice and video services continues to be a top issue as IPv6 becomes more widely used and algorithmic complexity increases. As part of a quantitative comparison approach, OPNET simulation was used to measure the primary QoS metrics in this triple redistribution scenario. The IS-IS addition had a favorable stabilizing impact, according to the results: The addition showed excellent stability in the IPv6 environment while maintaining great efficiency in the IPv4 context. The IP/MPLS core's efficient use of computational resources, which allowed QoS measurements to stay well within the Global QoS Acceptability Limits with the computing cost of utilizing three distinct routing algorithms. These findings effectively show the architectural viability of complex redistribution chains in next-generation networks while denying traditional worries about IPv6 routing complexity; this validates the ability of modern infrastructures to efficiently support multimedia services while reducing algorithmic overhead.

An Intelligent Monitoring System For DMVPN-Based Network

Dana Bassam Abdullah — 2026-09-30

This paper presents the design, implementation, and evaluation of a secure, scalable, and intelligent monitoring framework for Dynamic Multipoint Virtual Private Network (DMVPN) deployments. The proposed architecture adopts a layered model in which a Wide Area Network serves as the underlay, while DMVPN Phase 3 with multipoint Generic Routing Encapsulation (mGRE) tunnels protected by IPsec and IKEv2 forms the secure overlay. Enhanced Interior Gateway Routing Protocol (EIGRP) is deployed for dynamic route discovery and optimized spoke-to-spoke communication. A Python-based monitoring engine, built on Netmiko, collects operational telemetry through read-only SSH sessions, normalizes it into JSON, and applies rule-based detection logic against a healthy baseline. The system operates with a ten-second polling interval and is integrated with a centralized dashboard that visualizes tunnel status, routing adjacencies, and security associations. Experimental validation in GNS3 demonstrates that tunnel and routing failures are detected within a single monitoring cycle, achieving an estimated improvement of approximately 95% in fault detection speed compared to traditional manual approaches. The framework consolidates distributed CLI-based verification into a unified, automation-ready interface, supporting proactive enterprise network supervision.

Benchmarking Dimensionality Reduction Techniques for ViT Features

Hamdi Ahmed Jaber — 2026-09-30

Content-Based Image Retrieval (CBIR) systems face a major challenge today in the era of Edge AI. Vision Transformers (ViT) generate high-dimensional feature vectors (768 dimensions). This creates serious memory and pro-cessing problems for devices with limited resources. This empirical study tests three different ways to reduce these dimensions: Principal Component Analysis (PCA), Autoencoders (AE), and Product Quantization (PQ). These methods were applied to ViT features using a merged dataset of Oxford5k and Paris6k, containing 11,475 images. The study focuses heavily on real-world constraints like storage size, search latency, and accuracy.
To ensure the findings are statistically robust, all experiments were repeated 10 times. The results show a clear trade-off. Traditional linear methods like PCA lose too much accuracy under heavy compression, dropping to 0.4907 mAP at 64 dimensions. On the other hand, Autoencoders keep the semantic quality much better, reaching a mean of 0.5106 mAP. Meanwhile, Product Quantization proved to be highly storage-efficient. It achieved a 47.9x com-pression ratio while maintaining a competitive accuracy of 0.5087 mAP.
Based on these results, the paper suggests PQ-128 as a strong candidate for systems with limited storage. It provides a 24x compression ratio with a very small 0.56% drop in accuracy. The paper aims to help engineers choose the best compression method based on their specific device constraints. This guide is useful for deploying high-demand local applications, such as smart surveillance systems and high-speed visual search, on affordable hardware.

A Computer Aided Diagnostic System for Bone Fracture Detection in X-Ray Images

Almokhtar Almabrouk — 2026-09-30

Bone fractures are among the most common musculoskeletal injuries, yet X-ray interpretation remains susceptible to human error, particularly in regions facing radiologist shortages. This paper presents the development and clinical deployment of an end-to-end Computer-Aided Diagnosis (CAD) system for automated binary bone fracture detection. We systematically evaluate four custom Convolutional Neural Network (CNN) architectures and two transfer learning models (EfficientNetB0 and VGG16) on a public musculoskeletal X-ray dataset (Kaggle: ahmedashrafahmed/bone-fracture, version 1). To mitigate overfitting and improve robustness, an aggressive data augmentation pipeline was applied to the training set. Experimental results show that transfer learning significantly outperforms custom architectures, with EfficientNetB0 achieving the strongest diagnostic performance: 93.8% accuracy, 0.96 AUC-ROC, and a clinically critical 92.7% sensitivity. To translate these findings into practical clinical utility, the optimal model was integrated into a secure, lightweight Flask-based web application featuring real-time inference, strict input validation, UUID-based file anonymization, and automatic post-inference data deletion for compliance with patient-privacy standards.

A Domain Driven Design Based Reference Architecture for IoT Systems Engineering

Yousef Abusetta — 2026-09-30

The Internet of Things (IoT) paradigm continues to gain significant traction across diverse do-mains, primarily driven by its transformative and socio-economic benefits. However, engineering robust IoT systems remains a formidable challenge, largely due to the inherent dynamism, volatility, and uncertainty of their operational environments. To mitigate this architectural complexity, microservices have emerged as a viable alternative to traditional, rigid monolithic frameworks, offering a scalable mechanism to fulfill critical quality attributes. To maximize the efficacy of a microservices architecture, its technical boundaries must align precisely with domain-specific business requirements. Domain-Driven Design (DDD) addresses this need by strategically decomposing complex domains into modular, highly manageable subdomains that map directly to individual microservices. Furthermore, to minimize architectural coupling, enhance system integrability, and optimize device energy efficiency, Event-Driven Architecture (EDA) serves as a critical operational enabler. This paper proposes a comprehensive reference architecture for IoT systems engineering, systematically leveraging DDD and EDA principles to harmonize business logic with decoupled, event-driven microservices.

Virtual Reality in Education Across Cultures

Fathia Lahwal — 2026-09-30

Virtual Reality is increasingly used in education to support immersive, interactive, and experiential learning. However, its adoption is influenced not only by technological capabilities but also by cultural expectations, pedagogical traditions, infrastructural conditions, and ethical considerations. This narrative review examines how these dimensions interact across diverse educational con- texts and explores their implications for culturally responsive and responsible VR implementation. The review synthesised interdisciplinary literature identified through searches of Scopus, Web of Science, IEEE Xplore, SpringerLink, and Google Scholar, focusing primarily on publications from 2009 to 2024. Relevant literature addressing VR or immersive technologies in education and their cul- tural, pedagogical, technological, infrastructural, or ethical dimensions was the- matically organised and mapped to four research questions. The findings indicate that VR adoption and learning experiences are context-dependent rather than uni- versally consistent. Cultural frameworks can provide useful interpretive perspec- tives on learner expectations, interaction, autonomy, and collaboration, but should not be treated as deterministic predictors of individual behaviour. Peda- gogical alignment, institutional readiness, technological accessibility, educator preparedness, learner well-being, and responsible governance are also important considerations. Overall, the review supports a context-sensitive approach to VR in education that aligns cultural, pedagogical, technological, institutional, and ethical considerations.