Investigation and Development of Parallel Large Scale Record Linkage Techniques

记录链接旨在匹配大型（管理）数据库中相同实体（如客户或患者）的记录。拟议研究的结果将在效率、准确性和人工干预需求方面改进现有技术。通过实验研究和随机建模，将评估数据清理、标准化和链接的传统和新方法的性能。将研究属性值的统计依赖性的影响。将在高性能计算机和办公室个人电脑集群上实施、分析和评价使用集群来分块大型数据集的新方法以及包括交互作用项在内的预测模型。

Record linkage aims at matching records of the same entity (like customer or patient) in large (administrative) databases. The outcomes of the proposed research will improve current techniques in terms of efficiency, accuracy and the need for human intervention. Through experimental studies and stochastic modelling the performance of traditional and new methods for data cleaning, standardisation and linkage will be assessed. The effect of the statistical dependency of attribute values will be studied. New methods using clustering for blocking large datasets, and predictive models including interaction terms will be implemented, analysed and evaluated on high-performance computers and office-based PC clusters.