Programa de Pós-Graduação em Tecnologia da Informação
Permanent URI for this communityhttps://repositorio.ufrn.br/handle/123456789/27975
Browse
61 results
Search Results
Master Thesis Análise histórica de violações arquiteturais em sistemas Clojure: um estudo quantitativo longitudinal(Universidade Federal do Rio Grande do Norte, 2025-09-19) Silva, Welligton Miguel da; Barbosa, Eiji Adachi Medeiros; https://orcid.org/0000-0002-8286-0017; http://lattes.cnpq.br/8833409749475821; Kulesza, Uirá; http://lattes.cnpq.br/0189095897739979; Cirilo, Elder José Reioli; http://lattes.cnpq.br/4349166507859931Software architecture consists of fundamental decisions that guide the construction and evolution of a system. Over time, accumulated modifications throughout the software lifecycle can violate previously defined criteria, characterizing architectural degradation. The absence of a formal continuous evaluation process causes these problems to be detected reactively, often only after causing significant impacts. To mitigate these risks, static analysis tools can be adopted to validate architecture proactively, enabling constant monitoring. One example of a tool that follows this approach is Clj-Depend, which validates dependencies between modules in systems implemented using the Clojure programming language. In this context, this work aims to investigate the occurrence and evolution of these violations in systems implemented in Clojure. For this purpose, a quantitative longitudinal study based on software repository mining was conducted, involving five services from a multinational technology company, where they process on average more than eight hundred requests per minute. The results showed that all services presented violation accumulation correlated with code base growth, with concentration in few critical rules. Statistical analyses confirmed significant association between lines of code and architectural degradation when adequate governance is absent. Despite scope limitations, the study demonstrated that architectural degradation is monitorable and can be integrated into the team’s development workflow through automated processes. The developed protocol offers a systematic approach for continuous architectural auditing, contributing to more sustainable software evolution practices.Master Thesis Uso de técnicas de detecção automatizada de erros em datasets supervisionados para correção de rótulos oriundos de pipelines de aprendizado fracamente supervisionado(Universidade Federal do Rio Grande do Norte, 2025-09-22) Leal, Nalbert Gabriel Melo; Araújo, Daniel Sabino Amorim de; https://orcid.org/0000-0001-5572-0505; http://lattes.cnpq.br/4744754780165354; https://orcid.org/0000-0003-3178-7793; http://lattes.cnpq.br/2378995337056252; Santos, Araken de Medeiros; http://lattes.cnpq.br/8059198436766378; Menezes Neto, Elias Jacob de; https://orcid.org/0000-0002-1153-8899; http://lattes.cnpq.br/9152955193794784; Xavier Júnior, João Carlos; http://lattes.cnpq.br/5088238300241110The high cost of data labeling for training machine learning models has motivated the development of weakly supervised learning (WSL), however, this approach frequently introduces label noise, affecting model performance. Among WSL techniques, data programming (DP) stands out by utilizing noisy sources (such as heuristics and pre-trained models) to perform automated, low-cost data labeling, resulting in potentially inaccurate labels that impact the end-model’s performance. The objective of this work is to evaluate the impact of automatic label error detection techniques when integrated into the data programming pipeline. An experiment was conducted to identify the impact of label error detection on the performance and cost of the DP pipeline. The impact of each technique on performance was evaluated using the Matthews correlation coefficient (MCC) metric (collected from the evaluation of the pipeline produced end-model), and the cost was measured by the pipeline’s execution time. The results demonstrate that, in most cases, the application of detection techniques significantly degraded the end-model performance. Only 4% of the pipelines implementing detection showed a statistically significant performance improvement superior to the experiment baselines. These improvements, when they occurred, were isolated and accompanied by a high computational cost. It is concluded that DP pipelines without detection techniques demonstrated a better performance-cost trade-off, proving to be a more efficient approach.Master Thesis Assistente de busca: uma abordagem RAG para busca semântica em documentos textuais da Assembleia Legislativa do Rio Grande do Norte(Universidade Federal do Rio Grande do Norte, 2025-08-25) Lima, Saint Clair da Cunha; Araújo, Daniel Sabino Amorim de; https://orcid.org/0000-0001-5572-0505; http://lattes.cnpq.br/4744754780165354; http://lattes.cnpq.br/4820778376362087; Gurgel, André Morais; https://orcid.org/0000-0002-1925-8031; http://lattes.cnpq.br/7215356171529987; Menezes Neto, Elias Jacob de; https://orcid.org/0000-0002-1153-8899; http://lattes.cnpq.br/9152955193794784; Rêgo, Thais Gaudêncio do; https://orcid.org/0000-0002-6608-4900; http://lattes.cnpq.br/3166390632199101The unprecedented growth in the creation and persistence of unsctructured textual documents in public institutions poses challenges for efficient information retrieval and data analysis. This research addresses these challenges, by proposing a prototype of a search assistant using the Retrieval-Augmented Generation (RAG) approach, specifically applied to documents produced by Assembleia Legislativa do Estado do Rio Grande do Norte (Alern). The proposed system leverages Natural Language Processing (NLP) techniques, vector databases, and Large Language Models (LLMs) to enable semantic search and the generation of relevant content as answers to query inputs. The research introduces an architecture capable of retrieving document fragments based on semantic similarity. User-provided queries are processed and used to search content with contextual relevance, which is then synthesized into coherent and contextually appropriate responses through an LLM. The results of automated evaluations, using BERTScore and an LLM-based judge, demonstrate the system’s effectiveness in performing information retrieval based on user input—achieving 79% precision and 69% recall in BERTScore (values considered satisfactory in text generation scenarios), as well as an 84% performance rating in the LLM judge assessment. Evaluations conducted by human volunteers during the testing phase indicated an overall approval rate of 79%, whereas evaluations performed by users in production yielded 96.6% non-negative feedback. Being powered by the RAG approach, the proposed assistant not only reduces the cognitive load associated with the manual analysis of large document collections but also provides a scalable and adaptable solution for continuously evolving datasets. This research contributes to bridging the gap between the availability of public data and the generation of searchable information, aligning with goals of transparency and accessibility in the legislative environment.Master Thesis Engineering a permissioned blockchain architecture for CrossInstitutional EHR sharing(Universidade Federal do Rio Grande do Norte, 2025-08-28) Lindbergh, Vitor Gonçalez; Barroca Filho, Itamir de Morais; https://orcid.org/0000-0003-1694-8237; http://lattes.cnpq.br/1093675040121205; http://lattes.cnpq.br/9497507870891516; Pereira, Diego da Silva; https://orcid.org/0000-0001-6658-1854; http://lattes.cnpq.br/8773385390737695; Fontes, Ramon dos Reis; http://lattes.cnpq.br/0295503822342359The healthcare industry faces critical challenges in managing Electronic Health Records (EHRs), particularly regarding data security, interoperability, and patient privacy. Traditional centralized systems often lead to fragmented data storage, vulnerabilities, and inefficiencies that compromise care and integrity. This dissertation explores how blockchain technology—leveraging decentralization, immutability, and transparency—can enhance secure and interoperable EHR management. A systematic mapping study of 35 peer-reviewed papers from Scopus highlighted blockchain’s potential to improve interoperability, security, and patient control, while also identifying ongoing concerns with scalability, privacy, and regulatory compliance. In response, this work proposes a blockchain-based architecture that defines clear functional and non-functional requirements focused on security, interoperability, and consent management. The solution integrates smart contracts, consensus mechanisms, and a permissioned blockchain network to enable robust, auditable data exchange across healthcare institutions. A proof-of-concept (PoC) built using Hyperledger Fabric demonstrates secure EHR transactions, fine-grained access control, and patientdriven consent management. The PoC was validated against defined Quality of Service (QoS) metrics, confirming technical feasibility and alignment with system goals. This research contributes a viable architectural approach for secure and efficient EHR sharing and lays the foundation for future work on scalability, infrastructure integration, and regulatory alignment.Master Thesis Arquitetura baseada em IoT e lógica fuzzy para potencializar a utilização de recursos hídricos(Universidade Federal do Rio Grande do Norte, 2025-07-14) Oliveira, Francisco Fábio de; Immich, Roger Kreutz; https://orcid.org/0000-0003-2483-6382; http://lattes.cnpq.br/0535777592588490; Silva, Gustavo Girão Barreto da; http://lattes.cnpq.br/9491033611706611; Gomes, Rafael Lopes; https://orcid.org/0000-0001-7922-0695; http://lattes.cnpq.br/5212299313885086; Meneguette, Rodolfo Ipolito; http://lattes.cnpq.br/4981889689994770A crescente demanda por segurança alimentar e a escassez de recursos naturais apresentam desafios globais, especialmente na agricultura, que é responsável por uma parte significativa do consumo de água. No Brasil, a agricultura possui grande importância econômica, mas o uso intensivo de água exige práticas mais sustentáveis. Gerenciar a irrigação de forma adaptativa e eficiente é desafiador devido à complexidade das múltiplas variáveis ambientais. Isso reduz a eficiência do uso da água e impacta negativamente a produtividade agrícola. O IrrigaFlow é uma arquitetura modular que automatiza a irrigação utilizando IoT, lógica fuzzy e processamento distribuído. Consiste em três camadas: Módulo IoT, Borda da Rede e Nuvem, permitindo monitoramento e ajustes em tempo real com base em dados ambientais locais. Essa abordagem otimiza o uso da água e melhora a resposta às condições climáticas. Por meio de simulações conduzidas em ambiente controlado, foi possível avaliar a eficiência da arquitetura em cenários variados, destacando sua capacidade de ajustar o manejo hídrico às necessidades específicas de cada cultura e ambiente. O IrrigaFlow apresenta-se como uma alternativa promissora para enfrentar os desafios de segurança alimentar, reduzindo o desperdício de água e promovendo práticas agrícolas sustentáveis.Master Thesis Acompanhamento do planejamento estratégico baseado em processos: uma ferramenta integrada ao SUAP para instituições públicas de ensino superior(Universidade Federal do Rio Grande do Norte, 2025-08-07) Oliveira, Alikson Suel Costa de; Lima, Jean Mario Moreira de; https://orcid.org/0000-0002-2324-9365; http://lattes.cnpq.br/7467476735834560; http://lattes.cnpq.br/8234093727676086; Almeida, André Gustavo Duarte de; http://lattes.cnpq.br/1800962258138555; Barroca Filho, Itamir de Morais; https://orcid.org/0000-0003-1694-8237; http://lattes.cnpq.br/1093675040121205Public higher education institutions face challenges such as the fragmentation of information systems, the difficulty of integrating strategic planning data with budget execution, and the limitation of tools for consolidated analysis of resources. These factors make it difficult a clear and up-to-date visualization of spending, which affects transparency, financial control and efficiency in the management of public resources. Nowadays, budget control is, in many cases, carried out manually or with the help of isolated systems that do not offer a unified vision, making it difficult strategic decisions making by managers. To face these challenges, the developed work consists of a computational tool aimed at monitoring budget execution based on institutional strategic planning. This solution is integrated with the Unified Public Administration System (SUAP) and allows centralizing and automating processes such as budget reallocation, credit movement and expense requisition, providing greater control and alignment with the objectives defined in the Institutional Development Plan (PDI). The tool is being developed with technologies such as Python and Django, and its process includes stages of planning, coding, testing, validation and training of user servers. The consolidation of the tool highlights the potential of the solution to transform the way budget execution is managed in the institutional context. The expectation is to improve operational efficiency, with a reduction in the time and effort dedicated to budget control by allowing stakeholders to focus on strategic decision making. The processes centralization and automation – initially applied to expense requisition flows, movement and reallocation of credits – will help to reduce the dispersion of information between sectors by promoting greater consistency and standardization in the execution of resources. In line with this scenario, a survey conducted with users indicated a positive perception of the proposal and confidence in its applicability in the institutional context. Such evidence reinforces the tool’s potential to promote more efficient and transparent management of public resources, contributing to strengthening the institutional performance of IFRN, which is the object of the case study of this work and it presented as a solution with potential for replication in other federal educational institutions.Master Thesis Desenvolvimento de um Sistema para Apoio à Adoção de DevSecOps(Universidade Federal do Rio Grande do Norte, 2025-02-25) Pinto, Marcelo Martins; Fontes, Ramon dos Reis; http://lattes.cnpq.br/0295503822342359; Barroca Filho, Itamir de Morais; Souza, Rodrigo Rocha Gomes e; Immich, Roger KreutzThe Brazilian Judiciary, comprised of ninety-four Courts (source: CNJ, 12/01/2023) and their respective Information and Communication Technology departments, faces the crucial challenge of unifying its judicial systems. Despite the efforts of the National Council of Justice (CNJ), there is still a qualifiable disparity between administrative and support systems, manifested in the lack of specific regulations for their development. This diversity of solutions, combined with different personnel structures, technical capabilities and infrastructures, makes it difficult to control the applications in use, meet deadlines for development, distribution and implementation and, especially, software security, with regard to aspects of vulnerability analysis. To address these challenges, the proposal is to implement software that supports IT areas in adopting the DevSecOps methodology, integrating development, security and operations. This approach aims to break down information islands resulting from lack of sharing, distribute responsibilities and knowledge more efficiently, increase transparency in the IT area, improve quality and reduce development time, in addition to optimizing software maintenance throughout its life cycle. The development of the solution was based on a literature review to identify best practices, such as including security analysis from the initial phases of the development life cycle and tools, such as Burp Suite, for security verification. The proposed solution is a software that supports IT areas in controlling applications in use, their configurations and dependencies, in automating the provision of infrastructure, software version releases and, especially, in identifying vulnerabilities, identified by various types of analysis, such as SAST, DAST and SCA. The main contributions brought by this research aim to contribute to the adoption of the DevSecOps culture and systems security in the Regional Labor Court of the 21st Region, proposing an innovative solution, by enabling centralized access to information about applications, their dependencies and results of vulnerability analyses, which can be replicated in other public bodies, whether they are part of the Judiciary or not.Master Thesis Arquitetura para automação de configurações de rede utilizando infraestrutura como Código (IaC)(Universidade Federal do Rio Grande do Norte, 2025-03-27) Marinho, Luís Fhelipe Ribeiro Gomes Netto; Immich, Roger Kreutz; Pinheiro, Marcos César Madruga Alves; https://orcid.org/0000-0003-2483-6382; http://lattes.cnpq.br/0535777592588490; http://lattes.cnpq.br/0252077406219665; Venâncio Neto, Augusto José; Fontes, Ramon dos Reis; Meneguette, Rodolfo IpolitoThe demand for communication infrastructure is growing drastically, driven by the adoption of new technologies, with a focus on automation, error detection, and high availability. This shift is leveraged through increasingly programmable networks, mainly using technologies such as Software-Defined Networking (SDN) and methodologies like Infrastructure as Code (IaC), allowing networks to be managed as software. These processes require network assets to support protocols like NETCONF and telemetry data transmission using structured formats such as YAML. However, these new methodologies are not easily applied to a legacy equipment base, which lacks adequate support for these protocols. In this context, this work addresses the development of a solution for automating network assets in two complementary areas. First, it proposes an architecture for managing legacy network equipment using an IaC methodology, enabling the automation of routine tasks, fault detection, early problem identification, and configuration verification in a corporate production environment. The focus is on access network devices, as they are more numerous and geographically dispersed. Second, this work contributes to the field of network traffic analysis by constructing a dataset containing corporate network traffic data, serving as a foundation for network analysis and model training. The integration of these two areas enables not only the automated management of network assets but also more accurate assessment and prediction of network traffic conditions and performance. The results of this work showed that it was possible to identify and correct configuration inconsistencies in the assets, use routines to assist in network troubleshooting, and monitor the health of devices through their telemetry.Master Thesis Recomendação de produtos financeiros utilizando aprendizado de máquina(Universidade Federal do Rio Grande do Norte, 2024-12-17) Batista, Jonathan Jalles Silva; Araújo, Daniel Sabino Amorim de; https://orcid.org/0000-0001-5572-0505; http://lattes.cnpq.br/4744754780165354; http://lattes.cnpq.br/0333672178508920; Menezes Neto, Elias Jacob de; Rego, Thais Gaudêncio doRecommendation systems play a crucial role across various sectors, including finance, by offering personalized suggestions to customers based on their past preferences. In the financial and credit industries, such systems have the potential to revolutionize how institutions engage with customers, particularly through personalized recommendations for financial products like investments, insurance, and loans. This study analyzed 208,570 customer records from three types of financial insurance services to develop a solution capable of supporting marketing strategies for offering these services to the clients of a fintech. Clustering with k-means, tested with two to six clusters, revealed significant customer segmentation patterns. While the two-cluster configuration achieved the highest Silhouette Score (0.4169), the four-cluster approach provided more informative segmentation for strategic purposes. For predictive modeling, after initial tests and random hyperparameter search with 5-fold and 10-fold cross-validation, XGBoost and LightGBM achieved 82.0% recall and 80.5 % F1-score. LightGBM was selected for final evaluation on validation data due to cost-benefit considerations. When applied to the validation set, which had a significantly different insurance distribution from the training data due to covariate shift, the model’s performance dropped significantly to 43.1% recall and F1-score to 39.0%. The model performed best with Insurance C (63.7% recall and 60.0% F1-score) but struggled with Insurances A (45.8% recall and 26.7% F1-score) and B (2.6% recall and 4.8% F1-score). When trained on 80% of the combined training and validation dataset and validated on the remaining 20%, LightGBM showed substantial improvements for Insurances A and C, achieving recall scores of 83.3% and 81.6 and F1-scores of 83.0% and 77.6%, respectively. In all cases, the models struggled with Insurance B. Although the model’s performance with the combined dataset was significantly better, covariate shifts pose a notable challenge in developing the solution for the purpose of this study.Master Thesis Jaguar: a hierarchical deep reinforcement learning approach with transfer learning for StarCraft II(Universidade Federal do Rio Grande do Norte, 2024-11-07) Sá, Gabriel Caldas Barros e; Madeira, Charles Andryê Galvão; https://orcid.org/0000-0002-4460-2877; http://lattes.cnpq.br/2381782879955327; https://orcid.org/0000-0003-1164-0655; http://lattes.cnpq.br/9036202898138041; Araújo, Daniel Sabino Amorim de; Fernandes, Marcelo Augusto Costa; Corruble, VincentReal-Time Strategy Games are environments that generally simulate real military situations and present a series of challenges for the field of Artificial Intelligence, such as high complexity, large action and state spaces, partially observable maps, dealing with multiple units at the same time, and also the fact that tasks can be performed within the scope of micromanagement or macromanagement. In particular, Reinforcement Learning has stood out in the application and evolution of techniques capable of dealing with these challenges. A systematic review of the literature was then carried out to understand the state-of-the-art use of Deep Reinforcement Learning in Real-Time Strategy games. Some of the most relevant information raised were the use of StarCraft II as the main simulation environment, the need for more studies addressing macromanagement, the good performance of hierarchical architectures, and the importance of using action masking and transfer learning techniques, which can reduce the complexity of the problem and the computational cost. Given this and the challenges in this field, this work proposes to develop Jaguar, a hierarchical architecture capable of handling both macro and micromanagement, while requiring few resources during training. We implemented techniques for action and state shaping, invalid action masking through the state and the communication between hierarchical levels, different types of rewards, the DQN model for tactical decisions, and DDQN for strategic decisions. Two agents were trained in base scenarios and transfer learning was applied to two other scenarios. The results show that the agent achieved significant learning in the base scenarios and demonstrated promising performance in transfer learning scenarios. The proposed approach effectively handled the complexity of most tested StarCraft II scenarios while using minimal computational resources, though some limitations remain, leaving room for future improvements.