The application of learning sciences with technology has been shown to boost learner interactions, yet the potential of advanced tool, particularly those that leverage Natural Language Processing (NLP), still very much untapped in learning contexts. This paper speaks to this age-old problem of generating quality Multiple-Choice questions (MCQs) – a prevalent but time-consuming mode of assessment – via the suggested comprehensive comparison study of template-based AI solutions. The study contrasts general-purpose Large Language Models (LLMs) with specialized MCQ-focused AI programs. The scientific approach employed was quite stringent, where each of the software applications was benchmarked using a common dataset of text across varying levels of complexity and topic. Results indicate that general-purpose LLMs, especially DeepSeek and ChatGPT, consistently present higher performance and reliability, especially when processing complex textual content. Whereas specialized tools offer distinctive formatting options, they exhibit decreasing performance as texts become more complex and signify strong operation constriction at the free versions. Developing solid and effective distractors turned out to be a complicated task for all the tested tools. We conclude the paper by presenting a standardized assessment model, making evidence-based recommendations for developers and teachers, and suggesting ways to incorporate various AI capabilities into modern educational assessment effectively.

Maternal childbirth experiences are crucial indicators of health care quality, patient satisfaction, and emotional health. The increasing use of social media platforms, such as Facebook, provides a unique opportunity to examine public sentiment and narratives around maternity care. However, limited studies have employed deep learning (DL)–based sentiment analysis (SA) to comprehensively analyze childbirth experiences in low-resource environments. This study utilizes a hybrid technique that integrates unsupervised topic modeling with supervised DL sentiment classification to capture both thematic breadth and emotional tone of birthing experiences. A dataset of Facebook comments was preprocessed and analyzed using word frequency analysis, latent Dirichlet allocation (LDA) for thematic extraction, sentiment classification using convolutional neural networks (CNNs), and robustly optimized BERT pretraining approach (RoBERTa). Statistical correlations between sentiment polarity and hospital service variables were examined using the chi-square test. The word frequency analysis revealed significant themes such as maternal care, labor and childbirth experiences, informal childbirth discussions, spiritual beliefs, and personal childbirth stories. SA found a majority positive sentiment (5106 instances) with widespread sentiments of trust and joy, although considerable occurrences of fear (1,851), sadness (1,564), and anger (1,114) indicated traumatic delivery experiences. The CNN model outperformed RoBERTa, which had an AUC of 0.9988 and an accuracy of 87%. Statistical study (chi-square test, p < 0.0001) showed a significant correlation between sentiment polarity and hospital service variables, indicating the influence of treatment quality on patient views. This study emphasizes the significance of SA in assessing maternal health care experiences and improving hospital practices. The findings highlight the need of increased communication, empathetic midwifery care, and patient-centered approaches in addressing unfavorable childbirth experiences. This study provides policymakers with data-driven insights to improve maternal health care policies, increase patient experiences, and ensure valuable maternity services.

The increasing circulation of misinformation and fake news in online environments poses growing and unpedictable risks to tourism destinations and geosites, where accurate communication is essential for visitor safety, heritage protection, and destination reputation. This problem is further intensified by the rapid and unchecked spread of digital content. This study develops a supervised machine-learning approach to identify deceptive content in tourism-related texts by combining domain-specific keywords extracted from the abstracts of the GeoJournal of Tourism and Geosites with a pre-labelled fake-news corpus. The text preprocessing phase utilized comprehensive Natural Language Processing techniques to clean the textual data, eliminate noise, and ensure high-quality feature extraction. Using Natural Language Processing techniques and five classification algorithms (logistic regression, Support Vector Machine, boosted decision tree, decision forest, and neural network), the models were trained and evaluated within the Microsoft Azure environment. Results show that simpler and more interpretable models, particularly boosted decision trees and logistic regression, outperform neural networks on the sparse, tourism-filtered dataset, achieving high accuracy, precision, and F1 scores. The performance evaluation was strictly validated through comprehensive confusion matrices and Receiver Operating Characteristic curves to compare overall classifier efficiency. The findings highlight the vulnerability of geosites to misinformation, especially regarding natural hazards, geomorphological features, and heritage narratives, and demonstrate the potential of AI-based tools to support reliable communication and decision-making in tourism management. The study provides a foundation for future multilingual and geospatially enhanced misinformation-detection systems that can strengthen the resilience and sustainability of tourism destinations and geosites.

Standard retrieval-augmented generation (RAG) can perform poorly in AI for IT Operations (AIOps) settings because it is topology-blind. Basic RAG retrieves isolated, flat text snippets without enforcing structural or causal constraints, causing large language models to generate explanations that contradict the running system’s actual dependency structure. To address this gap, we conducted a systematic review following PRISMA 2020, searching Scopus, IEEE Xplore, Web of Science, and Google Scholar (last searched 31 January 2026). We included empirical or systems-oriented studies applying graph-based retrieval to ground a generative model in an IT, cloud, or software-operations setting, and excluded generic RAG without an operational context and graph-only methods without a generative component. Of 139 unique records, 31 met the criteria. Because reported metrics, tasks, and hardware were too heterogeneous for pooled effect estimates, we performed a descriptive quantitative synthesis of reporting frequencies for five outcome variables (localization accuracy, text/classification scores, MTTR, retrieval/inference latency, and graph construction cost), with values harmonized to common units. The synthesis reveals that hybrid-fusion approaches have become the dominant retrieval strategy, outpacing standalone traversal in adoption for Root Cause Analysis (RCA) tasks by fusing semantic vector search with strict structural constraints. However, our evaluation matrix exposes a critical production barrier: in 73% of Service Dependency Graph (SDG)-centric studies, topology drift or streaming-update handling is not explicitly described, with many pipelines evaluated on static or periodically refreshed snapshots. We outline Event-Sourced Streaming GraphRAG (ES-GraphRAG) as a reference architecture that frames concrete design requirements for latency and drift constraints based on an event-sourced, streaming construction pattern for snapshot-consistent retrieval. The framework also incorporates strict retrieval-time governance and budget-aware traversal to help keep LLM grounding topologically accurate and compliant with incident-response Service Level Objectives (SLOs).