Record Linkage Quality Without Ground Truth

In practice, the true matches in a linked dataset are rarely known. This project develops cluster-quality metrics that evaluate record-linkage output without ground truth, and scalable local-search algorithms that use these metrics to find and correct erroneous record assignments in noisy and incomplete data.

• “Using Cluster Quality Metrics for Record Linkage Evaluation,” ICCABS 2026 (forthcoming, Springer).
• Journal extension on unsupervised evaluation and refinement of record-linkage clusters, under review at the Journal of Computational Biology.