Modelado Semántico de Emergencias del ECU 911 con NLP y Ontologías

Abstract

This study proposes a novel hybrid framework for knowledge representation in emergencies, integrating Natural Language Processing (NLP), OWL ontologies, and SWRL rules to process unstructured data from Ecuador's Integrated Security Service (ECU 911). The key contribution lies in the unique combination of advanced NLP models such as BERT for Named Entity Recognition and XLM-RoBERTa for zero-shot semantic classification, with a formally validated ontological model developed in Protégé and a parallel logical implementation in Prolog using the Object-Attribute-Value paradigm. Unlike prior works, this approach specifically addresses the challenge of transforming raw emergency call transcripts into actionable knowledge by (1) automating entity extraction (locations, persons) and semantic categorization of incidents, (2) generating interpretable decision rules via decision trees, and (3) enabling cross-paradigm interoperability through synchronized OWL/SWRL and Prolog inference engines. Experimental validation with SPARQL/SQWRL queries and the Pellet reasoner demonstrated 96.7% accuracy in inferring emergency priorities such as medical emergencies, outperforming standalone NLP or ontology-based methods. This work advances semantic AI for emergency response by bridging unstructured text analysis with formal reasoning, offering a scalable solution for real-time decision support in critical scenarios.