Paper: Kernel-based Reranking for Named-Entity Extraction

ACL ID C10-2104
Title Kernel-based Reranking for Named-Entity Extraction
Venue International Conference on Computational Linguistics
Session Poster Session
Year 2010

We present novel kernels based on struc- tured and unstructured features for rerank- ing the N-best hypotheses of conditional random fields (CRFs) applied to entity ex- traction. The former features are gener- ated by a polynomial kernel encoding en- tity features whereas tree kernels are used to model dependencies amongst tagged candidate examples. The experiments on two standard corpora in two languages, i.e. the Italian EVALITA 2009 and the En- glish CoNLL 2003 datasets, show a large improvement on CRFs in F-measure, i.e. from 80.34% to 84.33% and from 84.86% to 88.16%, respectively. Our analysis re- veals that both kernels provide a compara- ble improvement over the CRFs baseline. Additionally, their combination improves CRFs much more than the sum of the indi- vidual contributions, su...