Federated query evaluation supported by SPARQL recommendation

Abstract

The usage of the semantic data is complicated for non-expert users, because the data are on many different SPARQL endpoints and the users need to know the URL of the endpoints. The federated systems are good solution for this problem. These systems interpret the SPARQL query and find the endpoints that can answer the query. Another advantage of the federated systems is it can connect the datasets with one simple query. This means some part of the data is found on different endpoints and the system can connect them. In order to do this the system needs information about the datasets that are stored by endpoints. We know two types for this. The first uses a catalog and stores the information in it. The system makes this information up-to-date at times. The second solution uses ASK query to decide which endpoint can answer the query. The queries run on every endpoint on every time. In this paper, we improve these federated systems. The basic idea is to reduce the number of the available endpoints. Our solution uses the SPARQL recommendation technique. This technique offers triple patterns to the user when the user writes the query. When the system asks new recommendation from the endpoints, it gets some information about the endpoints. This information is useful for the federated systems. It is enough to use only these endpoints given by the recommendations. In this paper, we extend the recommendation technique with new recommendation type that is based on the rdfs:range predicate. We present a query cost model that represents the number of the queries that need to the federated query evaluation. We present the recommendation information is enough for the evaluation and we make experiments on two federated systems (FedX, DarQ).