Infraestrutura para a governança de grafos de conhecimento: o caso da plataforma ELLAS

Abstract

Knowledge graph technology is developing rapidly, as it allows different types of data to be related in a semantic context, enabling complex analyses across many sectors. However, consolidated models in the literature for structuring such platforms treat the underlying infrastructure layer as a “black box”, without detailing how to ingest, triplify, store, and serve data sustainably under budget constraints. This gap imposes a technical and financial barrier on medium-sized social-impact projects: the absence of planned infrastructure can lead to slow performance, limited scalability, processing bottlenecks, and high operational costs, rendering initiatives such as ELLAS unfeasible. Against this backdrop, this research asks: how can a reference architecture be structured to enable, in a sustainable and low-cost manner, the technical governance of knowledge graphs in medium-sized projects such as ELLAS? Accordingly, the general objective is to propose a reference architecture for the governance of knowledge graphs in medium-sized, low-cost, and sustainable projects, instantiating it concretely in the ELLAS platform. The adequacy of the proposal is demonstrated by the performance observed under real load in a production environment, supported by cloud elasticity as a resource-provisioning mechanism. The proposed solution was applied and validated through the ELLAS (Equality in Leadership for Latin American STEM) Platform Case. The outcome of this work is twofold: (i) a reference architecture, generalizable to medium-sized knowledge-graph governance projects; and (ii) its concrete instantiation in the ELLAS platform, comprising the automated ingestion, triplification, and loading pipeline, with data exposed through a documented REST API (Swagger) in production. To support the solution, a Systematic Literature Review (SLR) focusing on infrastructures and semantic data integration was conducted. The results culminated in the modeling of an infrastructure based on the Onion Architecture pattern, separating the ontological domain from the orchestration and storage components. The architecture was initially prototyped in the cloud at low cost and subsequently deployed in an institutional production environment. Monitoring of the production environment, subjected to load during a launch event, confirmed the robustness of the solution, supporting concurrent access. Therefore, the proposed model enables the implementation of knowledge graphs in a sustainable and accessible manner, with adequate performance under real load, while horizontal scalability remains as future work — contributing to filling the technological gap in social-impact projects and democratizing the use of the Semantic Web.