Cyberbullying against women constitutes one of the most widespread forms of digital violence, manifesting through sexual harassment, non-consensual dissemination of intimate images, threats of physical or sexual aggression, defamation campaigns, and persistent offensive communications. This study conducts a systematic review based on PRISMA 2020, analyzing 31 studies published between 2015 and 2025. The review identifies key risk factors, including gender, age, socioeconomic status, time spent online, family environment, sexual orientation, and prior victimization experiences. These factors interact with contextual conditions such as patriarchal culture, school dynamics, and institutional barriers, which exacerbate women’s vulnerability. Quantitative synthesis shows that online harassment (87
Retrieval-Augmented Generation (RAG) has emerged as a key pattern to ground large language models (LLMs) on external knowledge sources. In practical deployments, these sources are frequently distributed across multiple tools, indexes, vector stores, and services that must be orchestrated over the network. The Model Context Protocol (MCP) provides a unified way to coordinate such heterogeneous components, but current architectures typically rely on uncoded, besteffort retrieval in which each retriever is queried separately, redundancy is manually configured, and failures are handled through ad hoc retries or fallback logic. This makes the pipeline sensitive to node failures, stragglers, bandwidth waste, and uneven answer quality. This paper extends the original NC-MCP-RAG proposal by integrating reviewer-requested considerations on Random Linear Network Coding (RLNC), coded caching, failure limits, overhead, and operational deployment, while grounding the design in previous work on distributed Fulcrum coding and AI-RMFgoverned zero-trust IoT architectures. We consider a design in which knowledge elements-documents, embeddings, or intermediate representations-are stored and exchanged in coded form, so that the orchestrator can reconstruct useful context from flexible subsets of responses. A prototype implementation is contrasted with sharded and centralized RAG configurations. The results show that network-coded retrieval can reduce median and tail latency, maintain answer quality under adverse conditions, and improve robustness under partial failures. The paper also discusses CPU and memory cost, cache dependence, privacy and governance constraints, extreme failure scenarios, and the architectural complexity introduced by coding.
Customer retention has become a strategic priority in the television and telecommunications sector, where companies increasingly rely on Business Intelligence and advanced analytics to identify churn risk and design targeted loyalty strategies. However, the intensive use of personal data, segmentation models, and automated decision-making processes raises significant ethical concerns related to privacy, transparency, fairness, and account-ability. This paper analyzes the ethical implications associated with a Business Intelligence system designed for customer retention in a subscription television company. The study examines how data collection, customer segmentation, predictive analytics, and automated decision mechanisms may affect data subject rights and fundamental ethical principles. Based on a conceptual and regulatory analysis, the research identifies the main ethical risks associated with such systems, including opacity of analytical processes, unfair decisions derived from inaccurate data, intrusive profiling, automated decision-making without human oversight, and excessive retention of personal data. As a main contribution, the paper systematizes the relationship between system components, ethical risks, and potentially affected rights, and proposes a conceptual framework of ethical safeguards integrating technical, organizational, and procedural measures to mitigate these risks. The proposed framework emphasizes explainability mechanisms, meaningful human oversight, data lifecycle management, and governance structures as key elements for responsible analytics. The findings provide practical guidance for organizations implementing Business Intelligence systems for customer retention and contribute to the broader discussion on ethical governance of data-driven decision systems in customer relationship management.
The COVID-19 pandemic accelerated the use of telemedicine and revealed persistent barriers in rural primary healthcare, especially among older adults and caregivers. This study proposes an LLM-mediated smart tele-primary-care framework to support respiratory infection monitoring in underserved communities of Córdoba, Colombia. The framework is based on caregiver-centered teleconsultation, remote monitoring, preventive education, structured symptom reporting, risk-based teletriage, and clinician-supervised digital support. The LLM functions as a controlled conversational interface to collect symptoms, organize patient information, reinforce health education, generate follow-up reminders, and identify predefined warning signs, without replacing clinical judgment or making autonomous diagnostic or treatment decisions. A preliminary scenario-based assessment examined whether the proposed workflow supports coherent triage and appropriate escalation of high-risk cases. Importantly, the described architecture—a telematic system for intelligent-engine access in assisted medicine support (ATIMAMS)—has been implemented and is currently operational at the Systems Engineering Program, Universidad del Sinú (Montería, Colombia), providing intelligent decision support for telemedicine consultations and remote assistance appointments in the study region. Overall, the study presents a context-sensitive, low-barrier, and safety-aware model for strengthening rural primary care, improving continuity of care, and supporting caregiver-mediated respiratory infection monitoring.
This paper presents a hybrid approach to facial emotion recognition that combines visual texture analysis via convolutional neural networks with structural features extracted from MediaPipe’s facial landmark mesh. Instead of applying trainable Graph Neural Networks, we define a graph-inspired representation where a subset of 30 expressive landmarks is encoded through geometric descriptors—specifically radial distances and angular orientations relative to the facial centroid. These features, concatenated with deep visual embeddings, enable robust and interpretable emotion classification across seven categories. The model operates in real time with average inference latency below 65ms per frame, requiring only standard CPU and webcam input. Experimental results on the FER2013 dataset demonstrate that incorporating graph-based descriptors improves class separability—particularly for high-arousal emotions—and enhances resilience to noise, occlusion, and head pose variation. This work offers an efficient and deployable alternative to full GNNs while preserving the benefits of topological facial reasoning.
