Studying Metadata for better client-server trade-offs in Linked Data publishing

Since the introduction of the Semantic Web, querying Linked Data mostly utilizes two types of interfaces: Linked Data documents, or the sparql protocol. However, both do not cover the wide spectrum of possible use cases and their specific requirements. Not only is the amount of public sparql endpoints is limited, they also suffer from frequent downtime [6, 15]. Predicting the consumption of computational resources of an endpoint is hard, because of sparql's expressiveness and individual user demand. Linked Data documents are more predictable, but querying based on traversing links is significantly slower and renders less complete results. Unfortunately, both are very undesired for reliable user applications. These issues above hint at a need for other client/server trade-offs. Such trade-offs can be analyzed using Linked Data Fragments (ldf) [17], which proposes an uniform view on all interfaces to rdf. A Linked Data Fragment is characterized by a specific selector links or uris to other fragments). This reveals a complete spectrum between Linked Data documents and the sparql protocol, in which we can advance the state-of-the-art of Linked Data publishing. This spectrum can be explored in the following two dimensions: i) selector, allowing different, more complex questions for the server; and ii) metadata, extending the response with more information clients can use. This work studies the second metadata dimension in a practical Web context. Considering the conditions on the Web, this problem becomes threefold. First, analog to the Web itself, ldf interfaces should exist in a distributed, scalable manner in order to succeed. Generating additional metadata introduces overhead on the server, which influences the ability to scale towards multiple clients. Second, the communication between client and server uses the http protocol. Modeling, serialization, and compression determine the extra load the overall network traffic. Third, with query execution on the client, novel approaches need to apply this metadata intelligently to increase efficiency. Concretely, this work defines and evaluates a series of transparent, interchangeable , and discoverable interface features. We proposed Triple Pattern Fragments (tpf) [17], a Linked Data interface with low server cost, as a fundamental base. This interface uses a single triple pattern as selector. To explore this research space, we append this interface with different metadata, starting with an estimated number of total matching triples. By combining several tpfs, sparql