Semantic Fusion for Connected Medicine

Abstract

The explosion in digital medical data makes it crucial to design intelligent assistants capable of interrogating this information reliably, quickly and contextually. In this article, we propose a novel comparative approach between two forms of knowledge representation: traditional tabular data and semantic ontologies. Based on the same clinical dataset concerning diabetic patients, we have implemented a dual structuring: a tabular version and an RDF ontology modelled with Protégé. An intelligent assistant, interfaced with the GPT-4 API, was designed to query both formats. The originality of our contribution lies in the experimental parallelisation of these two data models, through a standardised series of 300 questions, classified according to three levels of increasing complexity. This methodology enables us to objectively assess the robustness, responsiveness and inference capacity of each approach. The results are unequivocal: the ontology systematically outperforms the tabular format, with exact response rates ranging from 97% to 100%, compared with 34% to 81% for the tabular format. In addition, the ontological approach shows better tolerance of ambiguous queries and stability in semantic interpretation. Over and above performance, this study highlights the potential of knowledge graphs as an architectural foundation for future medical decision support systems. It also paves the way for hybrid systems that combine the accessibility of tables with the semantic power of ontologies - a perspective that has so far been little explored in the context of connected healthcare.