Scalable Record Linkage and Entity Resolution

Record linkage finds records across datasets that refer to the same person or entity. At scale, two costs dominate: comparing strings, and the huge number of candidate record pairs.

• Linear-time Jaro similarity. Standard algorithms for the Jaro similarity metric take quadratic time in the worst case. We gave an O(l₁ + l₂) algorithm that computes the same measure (ISBRA 2023; Journal of Computational Biology 2026).
• SuperBlocking. A blocking technique that removes unnecessary comparisons while keeping likely matches, with parallel record-linkage algorithms achieving near-linear speedups (IEEE BigData 2023).
• Phonetic and LLM-era blocking. Soundex and Double Metaphone blocking (IEEE BigData 2024; ISBRA 2025), and blocking for LLM-based record linkage (Algorithms 2025).

This work is in collaboration with researchers at the U.S. Census Bureau and the University of Florida.