Primary speaker:
Marlene van Ballegooie, Metadata Technologies Manager, University of Toronto Libraries
Description:
Academic libraries manage vast and complex collections in support of teaching and research, yet understanding how well these collections serve their communities remains a persistent challenge. Traditional assessment relies heavily on circulation counts and electronic resource usage statistics, measures that capture only a narrow slice of how people engage with library resources and reveal little about whether collections align with institutional research, the questions users are asking, or emerging areas of scholarly interest.
This session presents an AI-enabled method designed to close that gap. Rather than treating usage data in isolation, the project integrates faculty research output, user discovery behaviour and item use as three complementary lenses within a unified analytical framework. Using Google Gemini Pro, each dataset is classified against the Library of Congress classification system, creating a shared subject structure that enables meaningful cross-analysis. Faculty publications reveal where institutional scholarship is concentrated; search queries expose information needs users articulate; and circulation and e-resource data show what is actually consulted and borrowed. Bringing these perspectives together allows us to identify subject strengths, surface areas of unmet demand, and detect emerging research activity that might otherwise go unnoticed, offering a richer picture of how collections correspond to institutional knowledge production and user behaviour.
The presentation will walk through the methodology, from data preparation and prompt design to classification validation and cross-dataset analysis. It will also address practical limitations of using large language models as classifiers, including challenges around scale, consistency and the importance of human oversight. Connecting research, discovery and use, this framework demonstrates how AI can move libraries toward integrated, evidence-based collection intelligence.
