DOCUMENTARY LANGUAGES AND ARTIFICIAL INTELLIGENCE FOR INFORMATION RETRIEVAL

A RETRIEVAL-AUGMENTED GENERATION-BASED APPROACH

Abstract

Knowledge organization and effective information retrieval are challenges in digital scientific environments, particularly in specialized domains that require consistent terminological representation and advanced discovery mechanisms. In this context, this study proposes the SmartRetrieval model, which integrates the Retrieval-Augmented Generation architecture with language models to improve information retrieval in specialized databases. This is an applied study with a qualitative approach and a methodological character. As part of the proposal, a multilingual microthesaurus was developed from a selected corpus of the International Society for Knowledge Organization (brazilian chapter) papers to support semantic query expansion and terminological normalization. SmartRetrieval was evaluated using a corpus of 40 scientific articles and assessed through precision, recall, and F1-score metrics. The system employs a domain-oriented search strategy in Elasticsearch and uses Small Language Model-based re-ranking to increase the relevance of retrieved results. The findings demonstrate that the integration of documentary languages and generative models reduces noise, expands semantic coverage, and improves the effectiveness of information retrieval. Furthermore, they highlight the importance of specialized intellectual mediation in the construction and validation of conceptual structures, since the quality of subject representation directly influences the performance of retrieval systems. It is concluded that the proposed approach contributes to the improvement of information organization, representation, and retrieval processes, strengthening information mediation in digital scientific environments.