Comparison of machine learning methods for the sequence labelling applications

BİLGİN M., AMASYALI M. F.

Sinyal İşleme ve İletişim Uygulamaları Kurultayı, Malatya, Turkey, 16 - 19 May 2015 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/siu.2015.7129870
  • City: Malatya
  • Country: Turkey
  • Keywords: Conditional Random Fields, Sequence Labeling
  • Bursa Uludag University Affiliated: No

Abstract

In this study, on artificial data sets, it was compared condition random fields(CRF) and classical machine learning(CML) types. First part of this study, the performances of CRF and CML types were measured on artificial data sets. As the result of studies, CML types, except Naive Bayes, performanced higher than CRF. The success of NR and CRF is high when the outputs consist of one distribution, in other case it stays low. Besides in this study, it was evaluated the effect of education set size on success. The second study was made to test this situation.