Fairness in entity matching

Abstract

Entity matching (EM) has been one of the long-standing challenges in databases domain. Tra- ditionally, the approaches for EM focuses on developing matching algorithms and works on pairwise similarity between attributes. The main objective of this process is to obtain correct matches accurately. However, the current models lack any incorporation of fairness into them. In this paper, we try to formalize the notion of fairness with respect to our Entity Matching Process. We use state of the art unsupervised Entity Resolution model and illustrate the fair- ness metrics which can be associated with it. We modify the existing algorithm to incorporate fairness, while maintaining a reasonable trade-o_ with our EM results. Finally we report our results and see the trade-o_ with the state-of-the-art EM model and the increase in fairness on the benchmark datasets. Parts of our code can be found here- Link