Efficient Policy-Based Inconsistency Management in Relational Knowledge Bases

Real-world databases are frequently inconsistent. Even though the users who work with a body of data are far more familiar not only with that data, but also their own job and the risks they are willing to take and the inferences they are willing to make from inconsistent data, most DBMSs force them to use the policy embedded in the DBMS. Inconsistency management policies (IMPs) were introduced so that users can apply policies that they deem are appropriate for data they know and understand better than anyone else. In this paper, we develop an efficient “cluster table” method to implement IMPs and show that using cluster tables instead of a standard DBMS index is far more efficient when less than about 3% of a table is involved in an inconsistency (which is hopefully the case in most real world DBs), while standard DBMS indexes perform better when the amount of inconsistency in a database is over 3%.