基于遗传规划和主动学习的本体实例匹配

针对现有实例匹配方法存在的准确率和学习效率不高的问题,提出了一种新的基于遗传规划和主动学习的链接规则学习方法,并用于本体实例匹配。设计了更合理的链接规则表示,并针对链接规则的特点,对遗传规划的初始种群产生、适应度函数和进化算子进行了详细设计。提出了一种考虑样本相关性的主动学习采样策略,使得稀有样本被优先训练。实验结果表明,该方法不仅学习效率更高,而且能够学习出高质量的链接规则,取得了较好的本体实例匹配结果。

Genetic programming and active learning based ontology instance matching

Aim at existing instance matching methods'' problem of low precision and learning efficiency, this paper proposed a novel genetic programming and active learning based linkage rule learning method, which is applied to ontologies instance matching. According to the characteristics of a more reasonable expression of linkage rule, This method carried on a detailed design of the initial population, fitness function and evolutionary operator of genetic programming. Meanwhile, it proposed a correlation-aware active learning sampling strategy, which make the rare training samples is preferred. The experimental results show that the proposed method is not only has higher learning efficiency, but also can learn linkage rules of high quality, which achieve good ontology instance matching results.