HARNESSING TOPIC MODELING AND KNOWLEDGE GRAPHS FOR ENHANCED DISCOVERY IN HEALTH INTEROPERABILITY RESEARCH THROUGH RETRIEVAL-AUGMENTED GENERATION

Technologies today rarely function in isolation. Disparate data repositories and fragmented information systems necessitate seamless interoperability to support both clinical and research applications. This dissertation develops a computational framework for systematically evaluating interoperability standards that facilitate data exchange between research technologies and health systems.Rather than assessing interoperability through a single standard or methodology, this research introduces a structured knowledge model that categorizes interoperability concepts and cross-references them with relevant components, technologies, and terminologies. To achieve this, the study integrates artificial intelligence (AI) techniques, including natural language processing, machine learning, and graph analytics to extract, classify, and visualize interoperability relationships from health informatics literature.The framework applies unsupervised and semi-supervised learning methods, using AI techniques such as Latent Dirichlet Allocation (LDA), Non-Negative Matrix Factorization (NMF), and BERTopic to discover latent topic structures across thousands of biomedical publications. These topic models serve as the foundation for a domain-specific knowledge graph, allowing deeper semantic understanding and organization of interoperability research.To enhance the accuracy of concept extraction and entity recognition, the study creates fine-tunes a biomedical transformer model, Health Interoperability Research Bidirectional Encoder Representation from Transformers (HIResBERT), adapted for health interoperability. This model supports contextualized sentence embeddings and Named Entity Recognition tailored to interoperability use cases.The resulting AI-powered knowledge graph is operationalized through Graph-based Retrieval-Augmented Generation (GraphRAG) application, which facilitates semantic question answering and contextual discovery. Researchers can navigate complex interoperability relationships, uncover gaps, and retrieve literature mapped to relevant standards.Additionally, this research offers a structured mechanism for analyzing interoperability trends and delivers insights. By making complex standards and relationships more navigable, the framework enables researchers to effectively learn about and engage with interoperability concepts and technologies. By leveraging AI to automate the extraction, synthesis, and navigation of domain knowledge, this dissertation presents a scalable and reusable framework for advancing interoperability, clinical decision support, and health data integration within clinical and translational science.Ultimately, this work demonstrates how AI can be harnessed to advance literature discovery, knowledge modeling, and conceptual clarity across key areas of health informatics, laying the foundation for more transparent, data-driven, and integrative approaches to health research.