The rapid growth of scientific and technological publications has increased the demand for automated methods capable of extracting structured knowledge from unstructured textual data. This problem is particularly relevant for chemical and technological texts, where essential information is often represented through chemical reactions, physical formulas, and domain-specific terminology that standard natural language processing techniques handle poorly. This paper proposes a hybrid information extraction method that combines Named Entity Recognition (NER), a rule-based MiningNOUN algorithm, and Conditional Random Fields (CRF) to improve the identification of entities and relationships in domain-specific scientific texts. The proposed approach integrates statistical and ontological principles, enabling the recognition of substances, processes, physical quantities, and formally structured expressions that are typically missed by baseline NER models. The method was evaluated on a corpus of chemical and technological texts describing experimental procedures and reaction processes. The results show that the combined NER + MiningNOUN + CRF configuration significantly increases the coverage of extracted entities compared to a standard NER pipeline, allowing the system to capture information expressed in both natural language and formal notation. The extracted entities and relations are integrated into an ontological knowledge graph compliant with RDF/OWL standards and further applied within a Retrieval-Augmented Generation (RAG) architecture. The proposed method supports the development of more reliable knowledge graphs for intelligent scientific data processing and can be adapted to other technical domains with complex symbolic representations.