Entity Resolution (ER) is the process of identifying entity profiles that refer to the same real-world object—a complex challenge in data integration. ER is typically divided into two main sub-tasks: Filtering and Verification (also known as Matching). Filtering rapidly selects candidate pairs that are likely to match, thereby reducing the quadradic search space, while Verification examines each candidate pair in detail to determine whether they truly correspond to the same entity.
The first part of this talk explores filtering methods from two different filtering paradigms, Blocking and Nearest Neighbor Workflows, and introduces a systematic approach to benchmarking these methods in a fair way despite their highly different nature. The second part of the talk will focus on matching algorithms, which are currently dominated by Machine and Deep Learning based methods. Rather than evaluating these algorithms in isolation, their performance will be assessed under realistic settings by comparing their performance over candidate sets extracted by different filtering algorithms.