基于VS-Adaboost的实体对齐方法

针对现有实体对齐方法大多以本体模式匹配为基础，处理异构关联数据集间对齐关系存在局限性且实体链接缺失问题严重的现状，在分析关联数据语义的基础上，提出了一种独立于模式的基于属性语义特征的实体对齐方法，对关联数据集中实体属性根据语义标签特征及统计特征建模，并采用有监督的可变样本集VS-Adaboost算法实现分类器优化。实验结果表明，该方法的时间效率、准确率、查全率较高，F测度效果较好。
Abstract：Entity alignment is the key technology needed to realize knowledge graph construction, information integration and sharing. Most of the existing entity alignment approaches are based on ontology pattern matching, and have limitations when dealing with the alignment of heterogeneous linked open data(LOD), where the problem of missing entity links is serious. In this paper, a schema-independent entity alignment approach based on attribute semantic features is proposed. The entity attributes of the LOD are modeled according to their semantic label characteristics and statistical features. The supervised variable set VS-Adaboost algorithm is used to realize classifier optimization. The experiments executed on selected datasets show that compared with conventional methods the approach is more efficient and has better efficacy in terms of accuracy, recall and F measurement.