Computing Fine-grained Semantic Annotations of Texts

Semantic annotation extends the notion of metadata for accessing and sharing contents of different resources in different applications, such as Name Resolution, Information Extraction, Query Answering, and Regulation Analysis. Many ad hoc systems of semantic annotation have been developed, which have a major difference in ontologies used as the annotation reference. For example, KIM platform [1] takes the PROTON ontology; DBpedia Spotlight1 uses the DBpedia ontology; Magpie system2 has a predefined small ontology; and SPART3 uses the ontology generated on-the-fly from the text. In this sense, our approach is similar to SPART in that the reference ontology is constructed from a domain specific text because general ontology such as PROTON or DBpedia ontology has very low coverage on domain specific texts. For example, DBpedia Spotlight can hardly find occurrences on our examined corpora. Different from SPART, our approach is logic based instead of linguistic pattern based, which can benefit from the state-of-the-art ontology techniques.