Generating Lexicalization Patterns for Linked Open Data

The concept of Linked Data has attracted increased interest in recent times due to its free and open availability and the sheer of volume. We present a framework to generate patterns which can be used to lexicalize Linked Data. We use DBpedia as the Linked Data resource which is one of the most comprehensive and fastest growing Linked Data resource available for free. The framework incorporates a text preparation module which collects and prepares the text after which Open Information Extraction is employed to extract relations which are then aligned with triples to identify patterns. The framework also uses lexical semantic resources to mine patterns utilizing VerbNet and WordNet. The framework achieved 70.36% accuracy and a Mean reciprocal Rank value of 0.72 for five DBpedia ontology classes generating 101 lexicalizations.