Improving Record Linkage through Metaheuristic Optimization

Abstract

The exponential growth of digital data has amplified the importance of record linkage (RL), a fundamental task in data quality management that identifies and merges records referring to the same real-world entity. A critical step in RL is the blocking process, which reduces computational cost by partitioning records into candidate sets. The effectiveness of blocking depends on the choice of blocking keys (BKs), and poor selection can either increase complexity or degrade linkage quality. Since manual BK selection is costly and supervised approaches require labeled data that are often unavailable, recent research has focused on unsupervised optimization-based methods.In this study, we investigate two bio-inspired metaheuristic algorithms—the Whale Optimization Algorithm (WOA) and the Grey Wolf Optimizer (GWO)—for automatic blocking key selection. Both algorithms reformulate BK selection as a feature selection problem, where candidate subsets of keys are optimized using a wrapper-based evaluation function that balances Pair Completeness (PC), Reduction Ratio (RR), and    F-measure. WOA exploits the bubble-net hunting strategy of humpback whales, while GWO models the social hierarchy and cooperative hunting behavior of grey wolves, enabling both to effectively balance exploration and exploitation in high-dimensional search spaces.Experimental evaluations on multiple real-world datasets, including standard RL benchmarks and an Arabic dataset, demonstrate that WOA and GWO outperform traditional blocking strategies and achieve competitive performance compared to recent metaheuristic-based methods. Both approaches yield stable convergence, improved recall, and high reduction ratios, confirming their effectiveness and robustness in enhancing large-scale record linkage.