Path Analysis for Refining Verb Relations
Timothy Chklovski, Patrick Pantel · 2004
Link discovery is the process of identifying complex patterns from (multi)-relational data. The quality of link discovery outputs depends on the quality of the underlying data. In this paper, we discuss a method for refining multi-relational data. We treat the data as a graph and apply global link analysis to refine the graph. Specifically, we re-estimate the presence of a relation between a pair of nodes from the evidence provided by multiple indirect paths between the nodes. Our approach applies to a variety of relations: transitive symmetric, transitive asymmetric, and relations inducing equivalence classes. We present preliminary results on a semantic network called VERBOCEAN, which contains 22,306 relations between 3,477 verbs.