Closing surveillance detection gaps for emerging digital drug discourse with ontology-enhanced large language models

LLMs are touted as a solution to enhance digital drug surveillance but lack domain knowledge to perform these tasks effectively. Yet, when LLM architecture is enhanced by ontological frameworks, knowledge graphs, and retrieval-augmented-generation, we can evolve surveillance efforts to address a core issue contributing to surveillance lag—drug-slang and coded-language. Integrating these established methods offers a novel, linguistically grounded approach to improving the validity and responsiveness of digital drug surveillance.