A note on using the F-measure for evaluating record linkage algorithms
- 346 Downloads
Record linkage is the process of identifying and linking records about the same entities from one or more databases. Record linkage can be viewed as a classification problem where the aim is to decide whether a pair of records is a match (i.e. two records refer to the same real-world entity) or a non-match (two records refer to two different entities). Various classification techniques—including supervised, unsupervised, semi-supervised and active learning based—have been employed for record linkage. If ground truth data in the form of known true matches and non-matches are available, the quality of classified links can be evaluated. Due to the generally high class imbalance in record linkage problems, standard accuracy or misclassification rate are not meaningful for assessing the quality of a set of linked records. Instead, precision and recall, as commonly used in information retrieval and machine learning, are used. These are often combined into the popular F-measure, which is the harmonic mean of precision and recall. We show that the F-measure can also be expressed as a weighted sum of precision and recall, with weights which depend on the linkage method being used. This reformulation reveals that the F-measure has a major conceptual weakness: the relative importance assigned to precision and recall should be an aspect of the problem and the researcher or user, but not of the particular linkage method being used. We suggest alternative measures which do not suffer from this fundamental flaw.
KeywordsData linkage Entity resolution Classification Precision Recall Class imbalance
This paper was developed during discussions at the Isaac Newton Institute as part of the programme on Data Linkage and Anonymisation, July to December 2016 (https://www.newton.ac.uk/event/dla). We like to thank David Hawking and Paul Thomas for their advice on the use of the F-measure in information retrieval and Mark Elliot, Ross Gayler, Yosi Rinott, Rainer Schnell, and Dinusha Vatsalan for their comments during the development of this paper.
- Christen, P.: Data Matching—Concepts and Techniques for Record Linkage, Entity Resolution, and Duplicate Detection. Data-Centric Systems and Applications. Springer, Berlin (2012)Google Scholar
- Christen, P.: Preparation of a Real Temporal Voter Data Set for Record Linkage and Duplicate Detection Research. Technical Report, The Australian National University (2014)Google Scholar
- Christen, P., Vatsalan, D., Wang, Q.: Efficient entity resolution with adaptive and interactive training data selection. In: IEEE International Conference on Data Mining, pp. 727–732. Atlantic City (2015)Google Scholar
- Gutman, R., Sammartino, C., Green, T., Montague, B.: Error adjustments for file linking methods using encrypted unique client identifier (eUCI) with application to recently released prisoners who are HIV+. Stat. Med. 35(1), 115–129 (2016)Google Scholar
- Hand, D.J.: Measuring classifier performance: a coherent alternative to the area under the ROC curve. Mach. Learn. 77(1), 103–123 (2009)Google Scholar
- McCallum, A., Nigam, K., Ungar, L.H.: Efficient clustering of high-dimensional data sets with application to reference matching. In: ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp. 169–178. Boston (2000)Google Scholar
- Naumann, F., Herschel, M.: An introduction to duplicate detection. In: Synthesis Lectures on Data Management, vol. 3. Morgan and Claypool Publishers (2010)Google Scholar
- Newcombe, H.B.: Handbook of Record Linkage: Methods for Health and Statistical Studies, Administration, and Business. Oxford University Press Inc, New York (1988)Google Scholar
- Winkler, W.E., Yancey, W.E., Porter, E.H.: Fast record linkage of very large files in support of decennial and administrative records projects. In: Proceedings of the Section on Survey Research Methods, pp. 2120–2130. American Statistical Association (2010)Google Scholar