Optimizing large-scale biomedical ontology alignment through co-operative NSGA-II

Abstract

Since extant biomedical ontologies may have different approaches to model biomedical knowledge, the problem of semantic heterogeneity arises. As a key technique to address semantic heterogeneity, ontology matching intends to discover correspondences between semantically related entities of biomedical ontologies. Since the optimal objectives of ontology matching problem, i.e. precision and recall, are often in conflict with each other, it is difficult to find one perfect solution for all the decision-makers. In this work, we further propose a Co-operative NSGA-II (C-NSGA-II) to deal with large-scale biomedical ontology matching problems. In particular, C-NSGA-II divides the original population into multiple subpopulations, and each subpopulation is individually evolved, which aims to increase the diversity of individuals in the population to reduce the probability of early convergence. In addition, C-NSGA-II uses the External Archive (ExA) to collect the dominant individuals for each subpopulation, and on this basis, an adaptive resource allocation is presented to dynamically adjust the subpopulation's scale, which can improve the algorithm's converging speed. The experiment uses the Ontology Alignment Evaluation Initiative (OAEI)'s benchmark and anatomy track to test C-NSGA-II's performance, and the experimental results show that it outperforms OAEI's participants.