Reverse engineering of tree kernel feature spaces

Daniele Pighin, Alessandro Moschitti

Research output: Chapter in Book/Report/Conference proceedingConference contribution

15 Citations (Scopus)

Abstract

We present a framework to extract the most important features (tree fragments) from a Tree Kernel (TK) space according to their importance in the target kernelbased machine, e.g. Support Vector Machines (SVMs). In particular, our mining algorithm selects the most relevant features based on SVM estimated weights and uses this information to automatically infer an explicit representation of the input data. The explicit features (a) improve our knowledge on the target problem domain and (b) make large-scale learning practical, improving training and test time, while yielding accuracy in line with traditional TK classifiers. Experiments on semantic role labeling and question classification illustrate the above claims.

Original languageEnglish
Title of host publicationEMNLP 2009 - Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: A Meeting of SIGDAT, a Special Interest Group of ACL, Held in Conjunction with ACL-IJCNLP 2009
Pages111-120
Number of pages10
Publication statusPublished - 1 Dec 2009
Externally publishedYes
Event2009 Conference on Empirical Methods in Natural Language Processing, EMNLP 2009, Held in Conjunction with ACL-IJCNLP 2009 - Singapore, Singapore
Duration: 6 Aug 20097 Aug 2009

Other

Other2009 Conference on Empirical Methods in Natural Language Processing, EMNLP 2009, Held in Conjunction with ACL-IJCNLP 2009
CountrySingapore
CitySingapore
Period6/8/097/8/09

Fingerprint

Reverse engineering
Support vector machines
Information use
Labeling
Classifiers
Semantics
Experiments

ASJC Scopus subject areas

  • Computational Theory and Mathematics
  • Computer Science Applications
  • Information Systems

Cite this

Pighin, D., & Moschitti, A. (2009). Reverse engineering of tree kernel feature spaces. In EMNLP 2009 - Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: A Meeting of SIGDAT, a Special Interest Group of ACL, Held in Conjunction with ACL-IJCNLP 2009 (pp. 111-120)

Reverse engineering of tree kernel feature spaces. / Pighin, Daniele; Moschitti, Alessandro.

EMNLP 2009 - Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: A Meeting of SIGDAT, a Special Interest Group of ACL, Held in Conjunction with ACL-IJCNLP 2009. 2009. p. 111-120.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Pighin, D & Moschitti, A 2009, Reverse engineering of tree kernel feature spaces. in EMNLP 2009 - Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: A Meeting of SIGDAT, a Special Interest Group of ACL, Held in Conjunction with ACL-IJCNLP 2009. pp. 111-120, 2009 Conference on Empirical Methods in Natural Language Processing, EMNLP 2009, Held in Conjunction with ACL-IJCNLP 2009, Singapore, Singapore, 6/8/09.
Pighin D, Moschitti A. Reverse engineering of tree kernel feature spaces. In EMNLP 2009 - Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: A Meeting of SIGDAT, a Special Interest Group of ACL, Held in Conjunction with ACL-IJCNLP 2009. 2009. p. 111-120
Pighin, Daniele ; Moschitti, Alessandro. / Reverse engineering of tree kernel feature spaces. EMNLP 2009 - Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: A Meeting of SIGDAT, a Special Interest Group of ACL, Held in Conjunction with ACL-IJCNLP 2009. 2009. pp. 111-120
@inproceedings{a1512bb1c2b3421f8d7d3cbb84f61770,
title = "Reverse engineering of tree kernel feature spaces",
abstract = "We present a framework to extract the most important features (tree fragments) from a Tree Kernel (TK) space according to their importance in the target kernelbased machine, e.g. Support Vector Machines (SVMs). In particular, our mining algorithm selects the most relevant features based on SVM estimated weights and uses this information to automatically infer an explicit representation of the input data. The explicit features (a) improve our knowledge on the target problem domain and (b) make large-scale learning practical, improving training and test time, while yielding accuracy in line with traditional TK classifiers. Experiments on semantic role labeling and question classification illustrate the above claims.",
author = "Daniele Pighin and Alessandro Moschitti",
year = "2009",
month = "12",
day = "1",
language = "English",
pages = "111--120",
booktitle = "EMNLP 2009 - Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: A Meeting of SIGDAT, a Special Interest Group of ACL, Held in Conjunction with ACL-IJCNLP 2009",

}

TY - GEN

T1 - Reverse engineering of tree kernel feature spaces

AU - Pighin, Daniele

AU - Moschitti, Alessandro

PY - 2009/12/1

Y1 - 2009/12/1

N2 - We present a framework to extract the most important features (tree fragments) from a Tree Kernel (TK) space according to their importance in the target kernelbased machine, e.g. Support Vector Machines (SVMs). In particular, our mining algorithm selects the most relevant features based on SVM estimated weights and uses this information to automatically infer an explicit representation of the input data. The explicit features (a) improve our knowledge on the target problem domain and (b) make large-scale learning practical, improving training and test time, while yielding accuracy in line with traditional TK classifiers. Experiments on semantic role labeling and question classification illustrate the above claims.

AB - We present a framework to extract the most important features (tree fragments) from a Tree Kernel (TK) space according to their importance in the target kernelbased machine, e.g. Support Vector Machines (SVMs). In particular, our mining algorithm selects the most relevant features based on SVM estimated weights and uses this information to automatically infer an explicit representation of the input data. The explicit features (a) improve our knowledge on the target problem domain and (b) make large-scale learning practical, improving training and test time, while yielding accuracy in line with traditional TK classifiers. Experiments on semantic role labeling and question classification illustrate the above claims.

UR - http://www.scopus.com/inward/record.url?scp=79959209319&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=79959209319&partnerID=8YFLogxK

M3 - Conference contribution

SP - 111

EP - 120

BT - EMNLP 2009 - Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: A Meeting of SIGDAT, a Special Interest Group of ACL, Held in Conjunction with ACL-IJCNLP 2009

ER -