Intrinsic Property-based Taxonomic Relation Extraction from Category Structure

摘要

We propose a novel algorithm to extract taxonomic (or isa/instanceOf)relations from category structure by classifying each category link. Previous algorithms mainly focus on lexical patterns of category names to classify whether or not a given category link is an isa/instanceOf. In contrast, our algorithm extracts intrinsic properties that represent the definition of given category name, and uses those properties to classify each category link. Experimental result shows about 5 to 18% increase in F-Measure, compared to other existing systems.