DBPedia Live

DBPedia is a community project to extract structured data from Wikipedia in the form of an Open Knowledge Graph. DBPedia is served as linked data using Entity-Pages, REST APIs, or SPARQL endpoints. The latest version of DBPedia counts more than 470 millions of RDF triples¹.

DBPedia Live (DBL) a changelog stream continuously published to keep DBPedia replicas in-synch. A Synchronization Tool designed to consume the batches and update the local DBPedia copy is available [1].

DBL data are obtained by means of the DBpedia Live extraction framework [2] that produces extracted metadata with annotations from Wikipedia infoboxes. DBL is composed by three sub-streams carrying the entities to add, those to remove, and the triples to re-add upon propagating the deletion of an entity. Entities and triples make use of DBO types and properties.

DBL uses a pull-based mechanism. An update of DBL consists of four compressed NT files. Two main diff files, i.e., removed, and added, determine the insertion and deletion stream. Two further streams share files for clean updates that are optional to execute: reinserted, which corresponds to unchanged triples that can be reinserted, and clear, which prescribe the delete queries that clear all triples for a resource. Relevant metadata about the DBPedia Live stream corresponds to license, format and human-readable description.

DBPedia Live already provides RDF Data that make use of DBPpedia ontology (DBO), i.e., a cross-domain ontology, manually created from the most used Wikipedia info-boxes. The latest version of DBO covers 685 classes and 2795 different properties. It is a directed-acyclic graph, i.e., classes may have multiple super-classes, as required to map it to schema.org. Although edits are often done in bulk, this information is not present at the level of granularity the data are shared from DBL. Changes come unordered and, thus, we decided to ingest them using time-stamped triples similar to what C-SPARQL [4] CQELS [5] engines do.