Unrestricted Coreference Resolution via Global Hypergraph Partitioning

摘要

We present our end-to-end coreference resolution system, COPA, which implements a global decision via hypergraph partitioning. In constrast to almost all previous approaches, we do not rely on separate classification and clustering steps, but perform coreference resolution globally in one step. COPA represents each document as a hypergraph and partitions it with a spectral clustering algorithm. Various types of relational features can be easily incorporated in this framwork. COPA has participated in the open setting of the CoNLL shared task on modeling unrestricted coreference.