基于多目标蚁群算法的稳定参考点选择

Stable Reference Data Selection Based on Multi-objective Ant Colony Algorithm

为提高异构数据实体分辨的参考点选择稳定性,提出了基于多目标蚁群算法的异构数据实体分辨参考点稳定选择方法。异构数据的实体分辨可选取若干匹配的数据对象作为参考点,将不同数据集中的对象转换为到各自参考点的距离向量,依据空间结构的相似性进行实体分辨;为选择出较为稳定的参考点子集并且确保实体分辨的准确性,以实体分辨准确性与参考点集的稳定性最优为目标建立模型。以文中所提方法选择的参考点子集与通过集成Filter方法选择的参考点子集的扩展昆彻瓦指标值作为参考点子集稳定性评价指标,以F1值作为实体分辨准确性的评价指标;利用多目标蚁群算法求解该模型,同时将结合参考点之间距离的均值、方差和最大信息系数作为蚁群算法的启发式信息,提高了蚂蚁搜索较好解的能力,实现了参考点子集的优化选择。实验结果表明,该方法能够在实体分辨准确性和参考点子集稳定性方面达到较好的平衡。

In order to improve the stability of reference data selection for heterogeneous data entity resolution,a method of stable reference data selection for heterogeneous data entity resolution based on multi-objective ant colony algorithm is proposed.Heterogeneous data entity resolution can select numbers of matching the data objects as reference data,other objects are converted to distance vector based on the reference data for the entity resolution based on the spatial structure of similarity.In order to select more stable reference data and ensure the accuracy of entity resolution,the optimal model of the entity resolution accuracy and stability reference data is established.We choose extended Kuncheva evaluation index of the reference data of the proposed method and that of integrated Filter method selection as an optimization objective,with F 1 value as the other optimization objective.Using ant colony algorithm to solve this model,at the same time we combine with the reference data of the distance between the mean value,variance and coefficient of maximum information as heuristic information of ant colony algorithm,improving the ability of the ant searching better solutions and achieving the optimal selection of reference data.The experiment shows that the method can achieve a great balance between the accuracy of solid resolution and the stability of reference data.