Keyword Search on RDF Graphs: It Is More Than Just Searching for Keywords

In this paper, we propose a model for enabling users to search RDF data via keywords, thus, allowing them to discover relevant information without using complicated queries or knowing the underlying ontology or vocabulary. We aim at exploiting the characteristics of the RDF data to increase the quality of the ranked query results. We consider different dimensions for evaluating the value of results and achieving relevance, personalization and diversity. 1 Keyword Search on RDF Graphs Typically, data in knowledge bases is represented using the RDF model. In RDF, everything we wish to describe is a resource that may be a person, an institution, a thing, a concept, or a relation between other resources. The building block of RDF is a triple, which is of the form (subject, predicate, object). The RDF Schema (RDFS) language is used to introduce useful semantics to RDF triples. It provides a built-in vocabulary for asserting user defined schemas within the RDF model. This vocabulary can be used to specify URIs as being of a specific type (classes, properties and instances), to denote special relationships between URIs. The flexibility of the RDF data model allows the representation of both schema and instance information in the form of RDF triples. The traditional way to retrieve RDF data is through SPARQL, the W3C recommendation language for querying RDF datasets. SPARQL queries are built from triple patterns and determine the pattern to seek for; the answer is the part(s) of the set of RDF triples that matches this pattern. The correctness and completeness of answers of SPARQL queries are key research challenges. SPARQL is a structured query language that allows users to submit queries that may precisely identify their information needs, but require users to be familiar with the syntax, and the complex semantics of the language, as well as with the underlying schema or ontology. Moreover, this interaction mode assumes that users are to some extent familiar with the content of the knowledge base and also have a clear understanding of their information needs. On the other hand, today, there is a growing interest on a keyword-based search functionality to answer the search needs of users who are looking for specific information obtained by integrating numerous and heterogeneous sources. In addition, it is important for the user to be able to define different criteria on how all the resulting information could be ranked and returned to the user.