Extractive summarization via overlap-based optimized picking

Gaokun Dai*, Zhendong Niu

*Corresponding author for this work

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

1 Citation (Scopus)

Abstract

Optimization-based methods regard summarization as a combinatorial optimization problem and formulate it as weighted linear combination of criteria metrics. However due to inconsistent criteria metrics, it is hard to set proper weights. Subjectivity problem also arises since most of them summarize original texts. In this paper, we propose overlap based greedy picking (OGP) algorithm for citation-based extractive summarization. In the algorithm, overlap is defined as a sentence containing several topics. Since including overlaps into summaires indirectly impacts on salience, summary size and content redundancy, OGP effectively avoids the problem of inconsistent metric while dynamically involving criteria into optimization. Despite of greedy method, OGP proves above (1-1/e) of optimal solution. Since citation context is composed of objective evaluations, OGP also solves subjectivity problem. Our experiment results show that OGP outperforms other baseline methods. And various criteria proves effectively involved under the control of single parameter β.

Original languageEnglish
Title of host publicationWeb Information Systems Engineering – WISE 2017 - 18th International Conference, Proceedings
EditorsLu Chen, Athman Bouguettaya, Andrey Klimenko, Fedor Dzerzhinskiy, Stanislav V. Klimenko, Xiangliang Zhang, Qing Li, Yunjun Gao, Weijia Jia
PublisherSpringer Verlag
Pages135-149
Number of pages15
ISBN (Print)9783319687827
DOIs
Publication statusPublished - 2017
Event18th International Conference on Web Information Systems Engineering, WISE 2017 - Puschino, Russian Federation
Duration: 7 Oct 201711 Oct 2017

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10569 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference18th International Conference on Web Information Systems Engineering, WISE 2017
Country/TerritoryRussian Federation
CityPuschino
Period7/10/1711/10/17

Keywords

  • Citation-based extractive summarization
  • Non-decreasing submodular objective function
  • Overlap-based optimization

Fingerprint

Dive into the research topics of 'Extractive summarization via overlap-based optimized picking'. Together they form a unique fingerprint.

Cite this