Ontology-Based SPARQL Extension Ranker - Basic Implementation in the Context of OptiqueVQS and Comparison with Collaborative Filtering

This paper addresses the basic implementation and possible competitiveness of an ontology-based ranking method, that for a SPARQL query under construction, generates a ranking of likely extensions, based on a set of past queries. An extension's rank is its conditional probability of being seen in the set of past queries, where the condition is that its seen on a query with the same constraints up to the extension point. The context here is SPARQL queries constructed with the visual query system OptiqueVQS, so the types of extensions is limited to the ones it can make. A ranking method is needed to let end-users of OptiqueVQS faster locate wanted extensions, because the number of possible extensions gets quickly hard to manage relative to the size of the ontology used. OptiqueVQS is part of the EU project Optique - Scalable End-user Access to Big Data. Optique uses the paradigm of ontology-based data access (OBDA), empowering end-users with the ability to query after data in domain vocabulary and relations that they are familiar with. How much of an advantage would it be to use an ontology-based ranker in this ontological setting? Further enhanced versions of the ranker, can in different ways take the semantics in to an account to bias the rankings in ways other methods do not. To get an idea of the possible competitiveness of future enhanced versions, a collaborative filtering-based ranking method is implemented to compare against. These two methods are pitted against each other in an experiment where they are scored based on how high up the rankings certain intentionally removed extensions get.