Inferencing and Truth Maintenance in RDF Schema

Contrary to earlier reports in literature, exhaustive forward inferencing is a feasible approach for practical RDF. It is sufficiently fast and the increase in necessary storage size is sufficiently small to make it work. Benefits of this approach are low-cost design and implementation, and very cheap query answering, since this task is reduced to simple lookup without inferencing. A potential problem of exhaustive forward inferencing is how to deal with statement deletion (an aspect often ignored thus far): when a statement is removed, some of its consequences may also have to be removed. The naive approach is to simply recalculate the entire deductive closure of the RDF store. A more sophisticated approach is based on truth maintenance: it tracks all deductive dependencies between statements, and uses this to determine which other statements will have to be removed as a consequence of a single deletion. This approach has the additional advantage of having deductive dependencies available for other tasks, such as business logic and change tracking. We give a detailed algorithm for such truth maintenance for RDF(S), and we compare the performance of this algorithm with that of the naive recomputation approach.