An integrated deterministic and nondeterministic inference algorithm for sequential labeling

Yu Chieh Wu, Yue Shi Lee, Jie Chi Yang, Show Jane Yen

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

4 Scopus citations

Abstract

In this paper, we present a new search algorithm for sequential labeling tasks based on the conditional Markov models (CMMs) frameworks. Unlike conventional beam search, our method traverses all possible incoming arcs and also considers the "local best" so-far of each previous node. Furthermore, we propose two heuristics to fit the efficiency requirement. To demonstrate the effect of our method, six variant and large-scale sequential labeling tasks were conducted in the experiment. In addition, we compare our method to Viterbi and Beam search approaches. The experimental results show that our method yields not only substantial improvement in runtime efficiency, but also slightly better accuracy. In short, our method achieves 94.49 F (β) rate in the well-known CoNLL-2000 chunking task.

Original languageEnglish
Title of host publicationInformation Retrieval Technology - 6th Asia Information Retrieval Societies Conference, AIRS 2010, Proceedings
Pages221-230
Number of pages10
DOIs
StatePublished - 2010
Event6th Asia Information Retrieval Societies Conference, AIRS 2010 - Taipei, Taiwan
Duration: 1 Dec 20103 Dec 2010

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume6458 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference6th Asia Information Retrieval Societies Conference, AIRS 2010
Country/TerritoryTaiwan
CityTaipei
Period1/12/103/12/10

Keywords

  • L-regularization
  • machine learning
  • part-of-speech tagging
  • support vector machines

Fingerprint

Dive into the research topics of 'An integrated deterministic and nondeterministic inference algorithm for sequential labeling'. Together they form a unique fingerprint.

Cite this