Fine tuning features and post-processing rules to improve named entity recognition

Authors:
Óscar Ferrández;Antonio Toral;Rafael Muñoz
Affiliations:
Natural Language Processing and Information Systems Group, Department of Software and Computing Systems, University of Alicante, Spain;Natural Language Processing and Information Systems Group, Department of Software and Computing Systems, University of Alicante, Spain;Natural Language Processing and Information Systems Group, Department of Software and Computing Systems, University of Alicante, Spain
Venue:
NLDB'06 Proceedings of the 11th international conference on Applications of Natural Language to Information Systems
Year:
2006

Citing 6
Cited 0

Message Understanding Conference-6: a brief history

COLING '96 Proceedings of the 16th conference on Computational linguistics - Volume 1
A maximum entropy-based word sense disambiguation system

COLING '02 Proceedings of the 19th international conference on Computational linguistics - Volume 1
Named entity recognition using an HMM-based chunk tagger

ACL '02 Proceedings of the 40th Annual Meeting on Association for Computational Linguistics
Named Entity Extraction using AdaBoost

COLING-02 proceedings of the 6th conference on Natural language learning - Volume 20
Introduction to the CoNLL-2002 shared task: language-independent named entity recognition

COLING-02 proceedings of the 6th conference on Natural language learning - Volume 20
Named entity recognition through classifier combination

CONLL '03 Proceedings of the seventh conference on Natural language learning at HLT-NAACL 2003 - Volume 4

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper presents a Named Entity Recognition (NER) system for Spanish which combines the learning and knowledge approaches. Our contribution focuses on two matters: first, a discussion about selecting the best features for a machine learning NER system. Second, an error study of this system which lead us to the creation of a set of general post-processing rules. These issues are explained in detail and then evaluated. The selection of features provides an improvement of around 2.3% over the results of our previous system while the application of the set of post-processing rules provides an increment of performance which is around 3.6%, reaching finally 83.37% f-score.