Abstract

Semantic Web aims to make information on the web interoperable between machines. However, the information found on the web is unstructured, which makes it difficult for machines to process it. For this reason, this information is structured through the RDF format providing semantic meaning. This paper presents a structured approach for the extraction of RDF triples from unstructured text, with the objective of transforming raw text into semantically linked knowledge. The proposed process is organized in a pipeline, which includes: text reading and preprocessing, pronoun and contraction extension, sentence segmentation, triple extraction, and semantic entailment of entities and predicates. Large-scale language models (LLMs) are used for entity and relationship extraction, allowing for accurate and contextualized identification. The extracted entities are linked to their respective global identifiers (URIs) using Babelfy, while predicates are linked by querying Linked Open Vocabularies (LOV). Subsequently, the semantic similarity of the URIs is validated using the Levenshtein distance, in order to ensure a proper correspondence between text terms and semantic resources. Finally, an RDF graph is constructed according to Semantic Web standards. This approach combines natural language processing techniques, semantic reasoning and open data linking, achieving greater precision and cohesion in the generation of knowledge graphs. The results obtained show the feasibility of this approach for semantic integration tasks and automatic knowledge construction from unstructured texts. © 2025 IEEE.