Ranking Distributed Knowledge Repositories

Increasingly many knowledge bases are published as Linked Data, driving the need for effective and efficient techniques for information access. Knowledge repositories are naturally organised around objects or entities and constitute a promising data source for entity-oriented search. There is a growing body of research on the subject, however, it is almost always (implicitly) assumed that a centralised index of all data is available. In this paper, we address the task of ranking distributed knowledge repositories—a vital component of federated search systems—and present two probabilistic methods based on generative language modeling techniques. We present a benchmarking testbed based on the test suites of the Semantic Search Challenge series to evaluate our approaches. In our experiments, we show that both our ranking approaches provide competitive performance and offer a viable alternative to centralised retrieval.