Taxonomy beats corpus in similarity identification, but does it matter?

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

Abstract

We present extensive evaluations comparing the performance of taxonomy-based and corpus-based approaches on SimLex- 999. The results confirm our hypothesis that taxonomy-based approaches are more suitable to identify similarity. We introduce two new measures of evaluation that show that all measures perform well on a coarse-grained evaluation and that it is not always clear which approach is most suitable when a similarity score is used as a threshold. This leads us to conclude that the inferior performance of corpus-based approaches may not (always) matter.
Original languageEnglish
Title of host publicationInternational Conference Recent Advances in NLP 2015
EditorsG. Angelova, K. Bontcheva, R. Mitkov
PublisherINCOMA Ltd
Pages346-355
ISBN (Print)ISSN 1313-8502
Publication statusPublished - 2015
EventRecent Advances in NLP -
Duration: 7 Sep 20149 Sep 2015

Conference

ConferenceRecent Advances in NLP
Period7/09/149/09/15

Fingerprint

Taxonomies

Cite this

Lê, M. N., & Fokkens, A. S. (2015). Taxonomy beats corpus in similarity identification, but does it matter? In G. Angelova, K. Bontcheva, & R. Mitkov (Eds.), International Conference Recent Advances in NLP 2015 (pp. 346-355). INCOMA Ltd.
Lê, M.N. ; Fokkens, A.S. / Taxonomy beats corpus in similarity identification, but does it matter?. International Conference Recent Advances in NLP 2015. editor / G. Angelova ; K. Bontcheva ; R. Mitkov. INCOMA Ltd, 2015. pp. 346-355
@inproceedings{c69a0856d54b4385bda7dd9d790a6b67,
title = "Taxonomy beats corpus in similarity identification, but does it matter?",
abstract = "We present extensive evaluations comparing the performance of taxonomy-based and corpus-based approaches on SimLex- 999. The results confirm our hypothesis that taxonomy-based approaches are more suitable to identify similarity. We introduce two new measures of evaluation that show that all measures perform well on a coarse-grained evaluation and that it is not always clear which approach is most suitable when a similarity score is used as a threshold. This leads us to conclude that the inferior performance of corpus-based approaches may not (always) matter.",
author = "M.N. L{\^e} and A.S. Fokkens",
year = "2015",
language = "English",
isbn = "ISSN 1313-8502",
pages = "346--355",
editor = "G. Angelova and K. Bontcheva and R. Mitkov",
booktitle = "International Conference Recent Advances in NLP 2015",
publisher = "INCOMA Ltd",

}

Lê, MN & Fokkens, AS 2015, Taxonomy beats corpus in similarity identification, but does it matter? in G Angelova, K Bontcheva & R Mitkov (eds), International Conference Recent Advances in NLP 2015. INCOMA Ltd, pp. 346-355, Recent Advances in NLP, 7/09/14.

Taxonomy beats corpus in similarity identification, but does it matter? / Lê, M.N.; Fokkens, A.S.

International Conference Recent Advances in NLP 2015. ed. / G. Angelova; K. Bontcheva; R. Mitkov. INCOMA Ltd, 2015. p. 346-355.

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

TY - GEN

T1 - Taxonomy beats corpus in similarity identification, but does it matter?

AU - Lê, M.N.

AU - Fokkens, A.S.

PY - 2015

Y1 - 2015

N2 - We present extensive evaluations comparing the performance of taxonomy-based and corpus-based approaches on SimLex- 999. The results confirm our hypothesis that taxonomy-based approaches are more suitable to identify similarity. We introduce two new measures of evaluation that show that all measures perform well on a coarse-grained evaluation and that it is not always clear which approach is most suitable when a similarity score is used as a threshold. This leads us to conclude that the inferior performance of corpus-based approaches may not (always) matter.

AB - We present extensive evaluations comparing the performance of taxonomy-based and corpus-based approaches on SimLex- 999. The results confirm our hypothesis that taxonomy-based approaches are more suitable to identify similarity. We introduce two new measures of evaluation that show that all measures perform well on a coarse-grained evaluation and that it is not always clear which approach is most suitable when a similarity score is used as a threshold. This leads us to conclude that the inferior performance of corpus-based approaches may not (always) matter.

M3 - Conference contribution

SN - ISSN 1313-8502

SP - 346

EP - 355

BT - International Conference Recent Advances in NLP 2015

A2 - Angelova, G.

A2 - Bontcheva, K.

A2 - Mitkov, R.

PB - INCOMA Ltd

ER -

Lê MN, Fokkens AS. Taxonomy beats corpus in similarity identification, but does it matter? In Angelova G, Bontcheva K, Mitkov R, editors, International Conference Recent Advances in NLP 2015. INCOMA Ltd. 2015. p. 346-355