Search

Predicting which entities are likely to be mentioned in scientific articles is a task with significant academic and commercial value. For instance, it can lead to monetary savings if the articles are behind paywalls, or be used to recommend articles that are not yet available. Despite extensive prior work on entity prediction in Web documents, the peculiarities of scientific literature make it a unique scenario for this task. In this paper, we present an approach that uses a neural network to predict whether the (unseen) body of an article contains entities defined in domain-specific knowledge bases (KBs). The network uses features from the abstracts and the KB, and it is trained using open-access articles and authors’ prior works. Our experiments on biomedical literature show that our method is able to predict subsets of entities with high accuracy. As far as we know, our method is the first of its kind and is currently used in several commercial settings.

Maintaining an Open Dataset comes at an extra recurring cost when it is published in a dedicated Web interface. As there is not often a direct financial return from publishing a dataset publicly, these extra costs need to be minimized. Therefore we want to explore reusing existing infrastructure by enriching existing websites with Linked Data. In this demonstrator, we advised the data owner to annotate a digital heritage website with JSON-LD snippets, resulting in a dataset of more than three million triples that is now available and officially maintained. The website itself is paged, and thus hydra partial collection view controls were added in the snippets. We then extended the modular query engine Comunica to support following page controls and extracting data from HTML documents while querying. This way, a SPARQL or GraphQL query over multiple heterogeneous data sources can power automated data reuse. While the query performance on such an interface is visibly poor, it becomes easy to create composite data dumps. As a result of implementing these building blocks in Comunica, any paged collection and enriched HTML page now becomes queryable by the query engine. This enables heterogenous data interfaces to share functionality and become technically interoperable.

During amateur cycling training, analyzing sensor data in real-time would allow riders to receive immediate feedback on how they are performing, and adapt their training accordingly. In this paper, a solution with Semantic Web technologies is presented that gives such real-time personalized feedback, by integrating the data streams with domain knowledge, rider profiles {\&} other context data. This solution consists of a stream reasoning engine running on a low-end Raspberry Pi device, and a tablet app showing feedback based on the continuous query results. To demonstrate this in a static environment, a virtual training app is presented, allowing a user to simulate an amateur cycling training.

Functions are essential building blocks of any (computer) information system. However, development efforts to implement these functions are fragmented: a function has multiple implementations, each within a specific development context. Manual effort is needed handling various search interfaces and access methods to find the desired function, its metadata (if any), and associated implementations. This laborious process inhibits discovery, and thus reuse. Uniform, implementation-independent access is needed. We demo the Function Hub, available online at https://fno.io/hub: a Web application using a semantic interoperable model to map function descriptions to (multiple) implementations. The Function Hub allows editing and discovering function description metadata, and add information about alternative implementations. This way, the Function Hub enables users to discover relevant functions independently of their implementation, and to link to original published implementations.

This paper presents the idea and our work of extracting and reassembling a genealogical network automatically from a collection of biographies. The network can be used as a tool for network analysis of historical persons. The data has been published as Linked Data and as an interactive online service as part of the in-use data service and semantic portal BiographySampo—Finnish Biographies on the Semantic Web.

Route planning providers manually integrate different geo-spatial datasets before offering a Web service to developers, thus creating a closed world view. In contrast, combining open datasets at runtime can provide more information for user-specific route planning needs. For example, an extra dataset of bike sharing availabilities may provide more relevant information to the occasional cyclist. A strategy for automating the adoption of open geo-spatial datasets is needed to allow an ecosystem of route planners able to answer more specific and complex queries. This raises new challenges such as (i) how open geo-spatial datasets should be published on the Web to raise interoperability, and (ii) how route planners can discover and integrate relevant data for a certain query on the fly. We republished OpenStreetMap’s road network as “Routable Tiles” to facilitate its integration into open route planners. To achieve this, we use a Linked Data strategy and follow an approach similar to vector tiles. In a demo, we show how client-side code can automatically discover tiles and perform a shortest path algorithm. We provide four contributions: (i) we launched an open geo-spatial dataset that is available for everyone to reuse at no cost, (ii) we published a Linked Data version of the OpenStreetMap ontology, (iii) we introduced a hypermedia specification for vector tiles that extends the Hydra ontology, and (iv) we released the mapping scripts, demo and routing scripts as open source software.