Incomplete Databases: Missing Records and Missing Values

Data completeness is an essential aspect of data quality as in many scenarios it is crucial to guarantee the completeness of query answers. Data might be incomplete in two ways: records may be missing as a whole, or attribute values of a record may be absent, indicated by a null. We extend previous work by two of the authors [10] that dealt only with the first aspect, to cover both missing records and missing attribute values. To this end, we refine the formalization of incomplete databases and identify the important special case where values of key attributes are always known. We show that in the presence of nulls, completeness of queries can be defined in several ways. We also generalize a previous approach stating completeness of parts of a database, using so-called table completeness statements. With this formalization in place, we define the main inferences for completeness reasoning over incomplete databases and present first results.