Record Matching to Improve Data Quality

Data Quality is defined in [TB9SJ as fitness for use, which implies that quality is relative to the use of data. Problems with data quality tend to fall into two categories: inconsistency among systems and inconsistency with reality. Format/syntax, semantic and value inconsistencies are representative of inconsistency among systems whereas incorrect and missing values are representative of inconsistencies with reality. In this paper, we address the record matchIng problem which is related to value inconsistencies and incorrect or missing values. Inconsistencies related to duplicated or partially overlapping information among systems occur if changes in one system are not reflected in the other systems for various reasons such as bad design, lack of trust among systems, etc. The difficulties inherent in attempts to identify entities from different interoperating systems (as they independently evolve over time) that refer to the same real life entity are known as the record matching problem. This is a typical problem in multi-system organizations where data residing in diverse systems needs to be merged, either for assessing financial risks or for cutting down costs associated with various projects. The methodology presented in this paper unifies a variety of techniques addressing the record matching problem, which we are considering as a classification task. The techniques used are the