Personalized keyword search on large RDF graphs based on pattern graph similarity

The structure of the ever increasing large RDF repositories is too complex to allow non-expert users extract useful information from them. Keyword search is an interesting alternative but in the context of RDF graph data, where query answers are RDF graph fragments, it faces two major problems: The query quality answer problem and the result computation algorithm scalability problem. In this paper we focus on empowering keyword search on RDF data by exploiting personalized information. We propose an original approach which exploits the structural summary of the RDF graph to generate pattern graphs for the input keyword query. Pattern graphs are structured conjunctive queries and are seen as possible interpretations of the unstructured keyword query. Personalized information is represented as collections of profile graphs, a concept similar to pattern graphs. The ranking of the results is achieved by measuring graph similarity between the user profile graph and the generated pattern graphs. Novel similarity metrics have been introduced which consider intrinsic and extrinsic similarity and take into account both structural and semantic characteristics of the pattern and profile graphs. Effectiveness and efficiency experimental results show that our approach can tackle the two major problems that hinder the widespread use of keyword search on RDF data.