TY - GEN
T1 - Extractive summarization via overlap-based optimized picking
AU - Dai, Gaokun
AU - Niu, Zhendong
N1 - Publisher Copyright:
© 2017, Springer International Publishing AG.
PY - 2017
Y1 - 2017
N2 - 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.
AB - 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.
KW - Citation-based extractive summarization
KW - Non-decreasing submodular objective function
KW - Overlap-based optimization
UR - https://www.scopus.com/pages/publications/85031431861
U2 - 10.1007/978-3-319-68783-4_10
DO - 10.1007/978-3-319-68783-4_10
M3 - Conference contribution
AN - SCOPUS:85031431861
SN - 9783319687827
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 135
EP - 149
BT - Web Information Systems Engineering – WISE 2017 - 18th International Conference, Proceedings
A2 - Chen, Lu
A2 - Bouguettaya, Athman
A2 - Klimenko, Andrey
A2 - Dzerzhinskiy, Fedor
A2 - Klimenko, Stanislav V.
A2 - Zhang, Xiangliang
A2 - Li, Qing
A2 - Gao, Yunjun
A2 - Jia, Weijia
PB - Springer Verlag
T2 - 18th International Conference on Web Information Systems Engineering, WISE 2017
Y2 - 7 October 2017 through 11 October 2017
ER -