U.S. Semantic Technologies Symposium Series
Details about Session 2, chaired by Hande McGinty and Kathleen Jagodnik:
Presentation Title: LLM-Driven Knowledge Graphs: From Unstructured Text to Structured Insight Presenter: Dr. Alon Bartal
Abstract:
Large Language Models (LLMs) have transformed how we interact with unstructured text, yet modern AI systems continue to struggle with the structured, interconnected data that underpins real-world decision-making. This talk introduces a hybrid paradigm that bridges this gap by integrating LLMs with Knowledge Graphs (KGs) and Graph Neural Networks (GNNs) to create transparent, domain-grounded, and actionable AI systems. After outlining the limitations of relying solely on unstructured data or structured data pipelines, I demonstrate how LLMs can be used to extract entities and relations from scientific literature, social media, and semi-structured biomedical resources, converting them into the semantic triples that populate large heterogeneous KGs. These enriched graphs serve as the foundation for powerful graph-based learning and prediction. Through three case studies: (1) cancer risk prediction integrating genomic, socioeconomic, and environmental features; (2) early detection of unreported side effects of GLP-1 receptor agonist drugs using biomedical knowledge and social media signals; and (3) classification of drug mechanisms (etiological vs. palliative) via heterogeneous KGs combined with GNNs, the talk illustrates how hybrid LLM-KG architectures outperform conventional approaches and enable new types of reasoning. Together, these results highlight a path toward scalable, explainable, continuously updated AI systems capable of capturing both the richness of unstructured text and the precision of structured knowledge, with implications for advancing healthcare, pharmaceuticals, risk modeling, and scientific discovery.
Biography Dr. Alon Bartal:
Alon Bartal, Ph.D. is an Assistant Professor (with tenure) and Director of the Information Systems Program at Bar-Ilan University’s Graduate School of Business Administration. His research combines AI-based analytical models, complex network analysis, and semantic technologies, integrating Knowledge Graphs, Large Language Models, and graph mining, to advance computational health, biomedical informatics, and social network analytics. He has developed AI-driven methods for detecting childbirth-related PTSD from clinical narratives, modeled drug and gene mechanisms for precision medicine, and studied bias in clinical documentation. His work in the mental health domain has been recognized repeatedly, including being highlighted by the U.S. National Institutes of Health (NIH) on its official Science Update news page in both 2023 and 2024. Dr. Bartal’s research has earned multiple distinctions, including the Best Theoretical Paper Award at the HICSS 53rd Conference, and it has been published in leading venues including the IEEE Journal of Biomedical and Health Informatics, IEEE Transactions on Knowledge and Data Engineering, and Bioinformatics.