Entity Resolution
Overview
Entity resolution identifies duplicate records or links records across databases when they refer to the same person, organization, product, place, or other entity. Systems typically narrow the search with blocking or retrieval, then compare candidate pairs using string and attribute similarity, learned embeddings, language models, or structural constraints; some also use graphs or multimodal evidence. Recent work combines these methods with label-efficient training, budgeted model queries, and human review, and evaluates them across settings such as multilingual records, enterprise data, and historical sources. A central challenge is balancing recall against false merges: missed candidates cannot be recovered by a better matcher, while a single mistaken link can spread through a cluster, so evaluation must measure both pair matching and final clustering quality.
Latest papers
40 papers4 papers in the last four weeks, against 2 the four weeks before. 0.0% of all new papers.