Query-independent learning to rank RDF entity results of SPARQL queries

Abstract

RDF is a data model to represent structured data on the web. SPARQL is a query language for RDF data that returns exactly matching results. Number of these results may be very high. By rapid growth of web of data the need for efficient ranking methods for results of this kind of queries is increased. Because of exactly matching results in SPARQL queries, the focus is on the query independent features for ranking them. We use a learning to rank approach with four sets of query independent features to rank entity results of SPARQL queries over DBpedia. These features include: features extracted from RDF graph, weighted LinkCount, search engine based and information content of the RDF resource. We investigate the performance of individual features and the combination of them in learning to rank entity results. Experiments show that the complete feature set has the best performance in rankings. As an individual feature, the proposed information content of the RDF resource is a good choice based on its performance in ranking and the elapsed time for extracting this feature.