There are two well known technical issues when reasoning with ontologies that contain hundreds of thousands of classes/subclasses and where change happens frequently.
The first problem, materializing type information, takes far too much time. In some triple stores, materialization takes almost as long as loading the data. Once an ontology changes, the entire materialization process has to start over.
The second problem, optimizing a SPARQL engine for a reasoning triple store, is more challenging than just using SPARQL as a retrieval language. In a non-reasoning SPARQL engine, optimizing is relatively straightforward, applying the right hash and sort joins once given the statistics of the database when it reorders appropriately. However, when SPARQL is used on top of a reasoner, suddenly additional considerations are required. In practice, you only know the statistics of each clause after you have done the reasoning.
This Webinar will discuss a new solution that mitigates or nearly solves both problems. We will discuss some indexing techniques that do not require materialization and we will cover how an ordinary backtracking technique can be very fast with the right reordering.
View a recording of the event here.
Download the presentation slides here.