Feature combination for sentence similarity

Ehsan Shareghi, Sabine Bergler

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

3 Citations (Scopus)


The possible combinations of features traditionally used for sentence similarity amount to a very large feature space. Considering all possible combinations and training a support vector machine on the resulting meta-features in a two step process significantly improves performance. The proposed method is trained and tested on the SemEval-2012 Semantic Textual Similarity (STS) Shared Task data, outperforming the task's highest ranking system.

Original languageEnglish
Title of host publicationAdvances in Artificial Intelligence - 26th Canadian Conference on Artificial Intelligence, Canadian AI 2013, Proceedings
Number of pages12
Publication statusPublished - 26 Sep 2013
Externally publishedYes
EventCanadian Conference on Artificial Intelligence 2013 - Regina, Canada
Duration: 28 May 201331 May 2013
Conference number: 26th

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume7884 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


ConferenceCanadian Conference on Artificial Intelligence 2013


  • Feature Combination
  • Feature Selection
  • Sentence Similarity

Cite this