PhD Dissertation Defense by WANG Sha | Peer Group Analysis using Knowledge Graphs and Large Language Models for Business Optimization

ABOUT THE TALK

Comparing similar entities, or peers, is central to decision-making over complex and restricted information, yet doing it reliably remains hard. It requires three linked capabilities: finding the right peers, matching up data that is organized differently across sources, and enabling analysis when the underlying data cannot be shared.

This dissertation presents a unified framework addressing all three. First, a knowledge-graph-based exploration system finds meaningful peers by connecting specific entities through shared concepts, letting analysts zoom out to broader categories or zoom in to supporting evidence. Second, a matching approach uses large language models to reconcile differently structured data sources, handling complex real-world cases where simple similarity fails. Third, a privacy-safe framework generates realistic synthetic data guided only by summary statistics, so analysis can proceed without ever exposing real records.

Together, these components turn peer analysis from isolated tasks into a coherent workflow, making complex information findable, comparable, and shareable.

ABOUT THE SPEAKER

Sha's dissertation is focused on structured methods for analyzing and comparing complex, restricted datasets — combining data mining, knowledge graphs, and LLM-driven automation to make heterogeneous information easier to search, align, and share safely. Her research interests center on data mining and automation: building systems that reduce manual effort in data discovery, integration, and preparation, particularly where data is messy, siloped, or privacy-sensitive. Leisure activities she enjoys include: reading (mostly non-fictions), sports and history.