Data quality improvement by constrained splitting

In the setting of relational databases, the schema of the database provides a context in which the data should be interpreted. As a consequence, the quality of a relational database depends strongly on the assumption that data fits this context description. In this paper, we investigate the case where the information provided by an attribute value exceeds the framework provided by the schema. It is shown that such an information overflow can have two orthogonal causes: (i) data about multiple attributes are jointly stored as one attribute and (ii) data about multiple tuples are jointly stored as one tuple. Needless to say, such erroneous information storage deteriorates the quality of the database. In this paper, it is investigated how data quality can be improved by a split operator. The major difficulty hereby is to take into account the constraints that are present in a relational database. A generic algorithm is provided and tested on the well-know Cora dataset