关键词: 实体分辨, 记录对划分, 粗糙集

Abstract: The present approaches of entity resolution vary in effectiveness and efficiency, normal record pairs and ambiguous record pairs are separated, so that different approaches can be applied to them. As to the record pairs to be partitioned, variable precision neighborhood rough set is used to compute the lower and upper approximation of similar record pairs and dissimilar record pairs respectively, to get the approximation sets and boundary region of all record pairs, and those record pairs in the boundary region are regarded as ambiguous, the rest are normal. How the thresholds of inclusion degree and distance in the variable precision neighborhood rough set affect the effectiveness of data partition is analyzed. Experiments are conducted to compare the accuracy of the normal, ambiguous and original record pairs while using similarity threshold and KNN to resolute, showing the effectiveness of partition.

Key words: entity resolution, record pair partition, rough set