A new design of ensemble classifiers for high-dimension entity resolution

For high-dimension Entity Resolution (ER), a new design of ensemble classifiers based on feature selection is proposed, which regards ER as a binary classification problem. Binary classifier’s classification performance and similarity measurement are defined, Support Vector Machine is adopted as the base classifier. Classification accuracy, output dissimilarity of classifiers and feature subset’s number are used as optimization objects. Feature similarity vector of two candidate records is calculated as input data. Based on ER’s characteristics, the multiobjective problem is translated into a single objective optimization and graph-based ant colony optimization to solve it. The proposed method is validated by high-dimension datasets. • Technologies for IQ improvement and assurance➝Data scrubbing and cleansing Additional