RDF data has become a vital source of information for many applications. In this thesis, we present a set of models and algorithms to effectively search large RDF knowledge bases. These knowledge bases contain a large set of subjectpredicate-object (SPO) triples where subjects and objects are entities and predicates express relationships between them. Searching such knowledge bases can be done using the W3C-endorsed SPARQL language or by similarly designed triple-pattern search. However, the exact-match semantics of triple-pattern search might fall short of satisfying the users’ needs by returning too many or too few results. Thus, IR-style searching and ranking techniques are crucial.

This thesis develops models and algorithms to enhance triple-pattern search. We propose a keyword extension to triple-pattern search that allows users to augment triple-pattern queries with keyword conditions. To improve the recall of triple-pattern search, we present a framework to automatically reformulate triple-pattern queries in such a way that the intention of the original user query is preserved while returning a sufficient number of ranked results. For efficient query processing, we present a set of top-k query processing algorithms and for ease of use, we develop methods for plain keyword search over RDF knowledge bases. Finally, we propose a set of techniques to diversify query results and we present several methods to allow users to interactively explore RDF knowledge bases to find additional contextual information about their query results.

Details