DBpedia Spotlight at the MSM2013 Challenge

DBpedia Spotlight [5] is an open source project developing a system for automatically annotating natural language text with entities and concepts from the DBpedia knowledge base. The input of the process is a portion of natural language text, and the output is a set of annotations associating entity or concept identifiers (DBpedia URIs) to particular positions in the input text. DBpedia Spotlight provides programmatic interfaces for phrase recognition and disambiguation (entity linking), including a Web API supporting various output formats (XML, JSON, RDF, etc.) The annotations generated by DBpedia Spotlight may refer to any of 3.77 million things in DBpedia, out of which 2.35 million are classified according to a cross-domain ontology with 360 classes. Through identity links, DBpedia also provides links to entities in more than 100 other languages, and tens of other data sets. This paper describes our application of DBpedia Spotlight to the challenge of extracting Person (PER), Location (LOC), Organization (ORG) and Miscellaneous (MISC) entities from microposts (e.g. tweets) as part of the MSM2013 Challenge at WWW2013. All of the code used in this submission is available as Open Source Software, and all of the data used is shared as Open Data. A description of the software, data sets and more detailed evaluations are available from our supporting material page at http://spotlight.dbpedia.org/research/msm2013/.