Use of Support Vector Machines in Extended Named Entity Recognition

Abstract

This paper explores the use of Support Vector Machines (SVMs) for an extended named entity task. We investigate the identification and classification of technical terms in the molecular biology domain and contrast this to results obtained for traditional NE recognition on the MUC-6 data set. Furthermore we compare the performance of the SVM model to a standard HMM bigram model. Results show that the SVM utilizing a rich feature set of a ±3 context window and POS features (MUC-6 only) had a significant performance advantage on both the MUC-6 and molecular biology data sets.