AbstractiveSurvey of text in today timef

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/305912913
A survey on abstractive text summarization
Conference Paper · March 2016
DOI: 10.1109/ICCPCT.2016.7530193
CITATIONS
62
READS
8,156
2 authors:
Some of the authors of this publication are also working on these related projects:
Text summarisation View project
human action recognition View project
N. Moratanch
Anna University, Chennai
5 PUBLICATIONS 166 CITATIONS
SEE PROFILE
Chitrakala Gopalan
Anna University, Chennai
125 PUBLICATIONS 481 CITATIONS
SEE PROFILE
All content following this page was uploaded by N. Moratanch on 07 November 2017.
The user has requested enhancement of the downloaded file.

2016 International Conference on Circuit, Power and Computing Technologies [ICCPCT]
A Survey on Abstractive Text Summarization
N. Moratanch
Research Scholar,
Department of Computer Science and Engineering,
CEG, Anna University,
Chennai, India
tancyanbil@gmail.com
Dr. S. Chitrakala
Associate Professor
Department of Computer Science and Engineering,
CEG, Anna University,
Chennai, India
au.chitras@gmail.com
Abstract—Text Summarization is the task of extracting salient
information from the original text document. In this process, the
extracted information is generated as a condensed report and
presented as a concise summary to the user. It is very difficult for
humans to understand and interpret the content of the text. In
this paper, an exhaustive survey on abstractive text
summarization methods has been presented. The two broad
abstractive summarization methods are structured based
approach and semantic based approach. This paper collectively
summarizes and deciphers the various methodologies, challenges
and issues of abstractive summarization. State of art benchmark
datasets and their properties are being explored. This survey
portrays that most of the abstractive summarization methods
produces highly cohesive, coherent, less redundant summary and
information rich.
Index Terms—Text Summarization, structure Based
Approach, semantic Based Approach, Sentence Fusion,
Abstraction Scheme, Sentence Revision, Abstractive Summary
I. INTRODUCTION
In recent times text summarization has gained its
importance due to the data overflowing on the web. This
information overload increases in great demand for more
capable and dynamic text summarizers. It finds the importance
because of its variety of applications like summaries of
newspaper articles, book, magazine, stories on the same topic,
event, scientific paper, weather forecast, stock market, News,
resume, books, music, plays, film and speech. Due to its
enormous growth, many topnotch universities like Aarhus
University-Denmark, National Centre for Text Mining
(NaCTeM)-Manchester University, etc. have been staunchly
working for its improvement.
As the volume of information and published data on the
World Wide Web is growing day by day, accessing and
reading the required information in the shortest possible time
are becoming constantly an open research issue. It is a tedious
task to gather all the information and then give the output in a
summarized form. Internet is a platform that fetches the
information from databases. But still this information is
massive to handle. So text summarization came into demand
that condenses the document into shorter version by
preserving the meaning and the content. A summary is thus
helpful as it saves time as and retrieves massive documents
data. Prior to this time, it was done by manual labour but
now-a-days automation has brought forth many advantages.
Text summarization approaches can be typically split into
two groups: extractive summarization and abstractive
summarization. Extractive summarization takes out the
important sentences or phrases from the original documents
and groups them to produce a text summary without any
modification in the original text. Normally the sentences are in
sequence as in the original text document. Nevertheless,
abstractive summarization performs summarization by
understanding the original text with the help of linguistic
method to understand and examine the text. The objective of
abstractive summarization is to produce a generalized
summary, which conveys information in a precise way that
generally requires advanced language generation and
compression techniques.
Abstractive summarization is an efficient form of
summarization compared to extractive summarization as it
retrieves information from multiple documents to create
precise summary of information. This has gained its popularity
due to the ability of developing new sentences to tell the
important information from text documents. An abstractive
summarizer displays the summarized information in a
coherent form that is easily readable and grammatically
correct. Readability or linguistic quality is an important
catalyst for improving the quality of a summary.
Fig. 1. Overview of Abstractive Summarization
This paper collectively summarizes the major methodologies
adapted, issues found, research and future directions in text
summarization. This paper is organized as follows. Section 2
depicts about the Structured based approach. Section 3
describes about Semantic Based approach. Section 4 depicts
Abstractive
Summarization
Approaches using prior
knowledge
(Structure based Approach)
Approaches using NLP
Generation
(Semantic based Approach)
978-1-5090-1277-0/16/$31.00 ©2016 IEEE

about Inferences made. Section 5 describes about Challenges
and future research directions. Section 6 depicts about
Experimental evaluation and finally this paper is concluded in
Section 7.
II .STRUCTURE BASED APPROACH
Structured primarily based approach encodes most vital
data from the document(s) through psychological feature
schemas like templates, extraction rules and alternative
structures like tree, ontology, lead and body, rule, graph based
structure. Completely different ways that are used in this
approach area unit are mentioned as follows.
Fig. 2. Overview of Structure based approach
A. Tree based method
This technique utilizes a dependency tree that represents
the text/contents of a document. Completely different
algorithms are used for content choice for outline e.g. theme
intersection algorithmic program or an algorithmic program
that uses native alignment try across of parsed sentences. The
technique uses either a language generator or associate degree
algorithm for generation of outline. Connected literature
victimization this methodology is as follows.
Regina Barzilay et al. [1] proposed a sentence fusion
technique that identifies the common information phrases by
using bottom-up local multi-sequence alignment. Sentence
fusion is a technique used in multigene summarization system.
In this approach, multiple documents are given as inputs and
the central theme is identified by processing those inputs using
theme selection and once the theme is finalized, they do
ordering for the sentences and this is done by using clustering
algorithm. Once the sentences are ordered, they are fused
using sentence fusion and the corresponding statistical
summary is generated.
B. Template based method
This technique uses a guide to represent a full document.
Linguistic patterns or extraction rules area unit are matched to
spot text snippets that may be mapped into guide slots. These
text snippets are the area unit indicators of the outline content.
Sanda M. Harabagiu et al [2] proposed both single and
multi-document summarization. They have adopted the
techniques that were presented in GISTEXTER for producing
both extracts and abstracts from the documents. GISTEXTER
is a summarization system implemented for information
extraction that targets the identification of topic-related
information in the input document and translates it into
database entries and later from these databases, the sentences
are added to the summary based on user requests.
C. Ontology based method
Many researchers have created effort to use the ontology
(knowledge base) to boost the method of summarization. Most
documents on the online are domain connected which leads to
same topic being discussed. Every domain has its own
information structure which is highly represented by ontology.
In the connected literature exploitation, this technique is
mentioned as follows.
Lee et al [3] proposed the fuzzy ontology with its ideas
introduced for Chinese news summarization to model
uncertain information and thus will precisely describe the
domain knowledge. In this approach, the domain ontology for
news events is outlined by the domain experts followed by the
Document preprocessing phase that produces the meaningful
terms from the news corpus and also the Chinese news
dictionary. For each of the fuzzy concept in the fuzzy
ontology, the fuzzy inference phase generates the membership
degrees. Various events of the domain ontology is associated
with the collection of membership degrees for every fuzzy
ideas.
D. Lead and body phrase method
This methodology relies on the operations of phrases
(insertion and substitution) that have same syntactic head
chunk within the lead and body sentences, so as to rewrite the
lead sentence.
Tanaka et al. [4] proposed a method for summarizing
broadcast news by analyzing the lead and body chunks of the
sentence syntactically. The baseline of this idea is inferred
from sentence fusion techniques. The summarization method
involves in identifying common phrases in the lead and body
chunks followed by insertion and substitution of phrases to
generate a summary of news broadcast through sentence
revision process. The initial step includes syntactically parser
of the lead and body chunks which are followed by identifying
trigger search pairs, followed by phrase alignment by using
different similarity and alignment metrics. The final step
involves insertion or substitution or both. The insertion step
Structure
Based
Approach
Tree
Based
Method
Template
Based
Method
Ontology
Based
Method
Lead and
Body
Phrase
Rule
Based
Method
Graph
Based
Method

involves decision of insertion point, redundancy check and
discourse coherence check to ensure coherency and
elimination of redundancy. The substitution step ensures to
increase information by substituting body phrase in the lead
chunk.
E. Rule based method
In this technique, the documents to be summarized
are depicted in terms of classes and listing of aspects. Content
choice module selects the most effective candidate among
those generated by data extraction rules to answer one or lot
of aspects of a category. Finally, generation patterns are used
for generation of outline sentences.
Pierre-Etienne et al. [5] suggested information
extraction rules to find semantically related nouns and verbs.
After extraction, content selection tries to avoid mixing
candidates and sends the data to the generation. It is used for
sentence structure and words in straight forward generation
pattern. After generating, content guided summarization is
performed.
Huong Thanh Le et al. [6] proposed an approach to
abstractive text summarization based on discourse rules,
syntactical constraints and word graph. The sentence reduction
step is based on input sentences, keywords of the original text
and syntactic constraints. Word graph is used only in the
sentence combination stage. The method of generating a
sentence from the essential fragment is split into finishing the
start of a sentence and finishing the tip of a sentence. Sentence
Combination is performed by observing and adhering to few
syntactical cases.
Ansamma John et al. [7] proposed text Summarization
based on feature score and random forest classification. The
given input is pre-processed and then it computes the feature
scores followed by training and cross validation of classifier
and finally generating the summary of required size by
maximal marginal relevance. The classification is a binary
problem that determines which class the sentence belongs to
either summary or non-summary class. The main task is to
generate summary sentences from the summary class. The
selected sentences are based on maximum relevance and
minimum redundancy.
F. Graph based method
Many researchers use a graph data structure (called
Opinosis-Graph) to represent language text. The novelty
introduced in the system is that every node represents a word
unit representing the structure of sentences for directed edges.
Dingding Wang et al. [8] suggested Multi-document
summarization systems based on a variety of strategies like the
centroid-based method, graph-based method, etc to evaluate
different baseline combination methods like average score,
average rank, borda count, median aggregation etc., for
achieving a consensus summarizer to improve the
performance of the summarization. A novel weighted
consensus scheme is proposed to collect the results from
individual summarization methods . Natural language
generation (NLG) system is fed using linguistics illustration of
document(s) in semantic based technique. This technique
specializes in identifying noun phrases and verb phrases by
linguistic data.
Kavita Ganesan et al. [9] proposed a novel graph-
based summarization framework (Opinosis) that generates
compact abstractive summaries of extremely redundant
opinions. It has some distinctive properties that are crucial in
generating abstractive summaries: Redundancy Capture,
Gapped ssubsequence ccapture, Collapsible sstructures. The
model generates an abstractive summary by exhaustively
searching the Opinosis graph for appropriate sub-graphs
encoding a valid sentence and high redundancy scores. The
major components of the system are meaningful sentence and
path scoring. Then a valid path is selected and it’s marked
with high redundancy score, collapsed paths and generation of
summary. Then all the paths are ranked in the descending
order of the scores and are eliminated duplicated paths by
using Jaccard measure.
Elena Lloret et al. [10] focused on generating abstract
summaries using word graph based method. The approach
combines both extractive and abstractive information to
generate abstracts. This method compresses and merges
information based on word graph method thus generating
abstracts. The words in the document form a set of vertices in
the graph and the edge that represents the adjacent relationship
between two words. A weighting function has been formulated
to define the importance of the threshold using Page Rank
value. The shortest path algorithm is used since it gives
minimal length sentence with more information from relevant
nodes in the graph. The important content is found using
Compendium Text Summarization approach through two
methods i) a set of sentences are given as input to the word
graph and then it is forwarded to the compendium ii) selecting
important content from the source document and then
applying word graph method.

TABLE I
A COMPARATIVE STUDY ON STRUCTURED BASED APPROACH
Author/
Year
Techniques/
methods
Text
representation
Content
selection
Summary
generation
Regina
Barzilay,
1999
Tree Based Dependency
Tree
Theme
intersection
algorithm
Sentence
fusion
Sanda
M.Harabagiu,
2002
Template
Based
Template
having slots
and
fillers
Linguistic
patterns or
Extraction
rules
IE based
summarization
algorithm
Lee and Jian,
2005
Ontology-
Based
Fuzzy
Ontology
Classifier News Agent
Tanaka and
Kinoshita,
2009
Lead and
Body Phrase
Lead, body and
Supplement
structure
Revision
Candidates
(Maximum
phrases of
some head
in lead and
body
sentences)
Insertion and
Substitution
operations
on phrases
Ganest and
Lapalme,
2012
Rule-Based Categories and
Aspects
Information
Extraction
rules
Generation
Patterns
Elena Lloret,
2011
Graph based Compresses
and merge
Word
Graph
Minimal
length
sentences
compression
III .SEMANTIC BASED APPROACH
In semantic based technique, linguistics illustration of
document(s) is employed to feed into natural language
generation (NLG) system. This technique specialize in
identifying noun phrases and verb phrases by processing
linguistic data.
Fig.3. Overview of Semantic based approach
A. Multimodal semantic model
In this technique, a linguistics model, that captures
concepts and relationship among ideas, is made to represent
the contents like text and images that are used for multimodal
documents. The important ideas are rated using some
measures and eventually the chosen concepts are expressed as
sentences to create summary.
Albert Gatt et al. [11] proposed a realization engine
that aims in building large-scale data-to-text NLG systems,
whose task is to summarize massive volumes of numeric and
symbolic data. SimpleNLG provides interfaces to offer direct
control over the way phrases are built and combined,
inflectional morphological operations, and linearization. The
major steps in constructing a syntactic structure and
linearizing it as text with SimpleNLG are initializing the basic
constituents that are required with the lexical items,
combining constituents into larger structures, passing the
resulting structure to the linearizer that traverses the
constituent structure by applying the correct inflections and
linear ordering depending on the features, and later the
realized string is returned.
B. Information item based method
In this methodology, the information about the
summary are generated from abstract representation of supply
documents, instead of sentences from supply documents. The
abstract illustration is Information Item that is the smallest part
of coherent information in a text.
Pierre-Etienne Genest et al. [12] generated
summarization by an information item (INIT) which is the
smallest element of coherent information in a text or a
sentence. The important goal is to identify all text entities,
their attributes, predicates between them, and the predicate
characteristics. During selection, the analysis of the source
documents that leads to a list of INIT will proceed to select
content for the summary. Frequency based models, such as
those used for extractive summarization, could be applied to
INIT selection instead of sentence selection. Most INIT do not
give rise to full sentences, and there is a need of combining
them into a sentence structure before being realized as a text.
Local decisions are designed how to present the information
to the reader and in what order during generation are now led
by global decisions of the INIT selection step. The final
summary generation is done by ranking of the generated
sentences and a number of sentences that are intentionally in
excess of size limit of the summary is first selected.
C. Semantic Graph Model
This technique aims to summarize a document by
creating a linguistics graph known as rich semantic graph
(RSG) for the initial document by reducing the generated
Semantic
Based
Approach
Information
Item Based
Method
Multimodal
Semantic
Method
Semantic
Graph Based
Method
Semantic
Text
Representati
on Model

linguistics graph and then generating the final abstractive
outline from the reduced linguistics graph.
Ibrahim et al. [13] summarized a single document by
creating a semantic graph called Rich Semantic Graph (RSG)
from the original document. Then it reduces the generated
semantic graph, and the final abstractive summary is produced
from the reduced semantic graph. In Rich semantic sub-graphs
generation module, for each input Word senses instantiation
process instantiates a set of word concepts for both verb and
noun senses based on the domain ontology. Concept validation
processes are interconnected and validated to generate
multiple rich semantic sub-graphs. The sentence concepts are
inter-linked through the syntactic and semantic relationships
generated in the pre-processing module. Sentence ranking
process aims to threshold the highest ranked semantic sub-
graphs for each sentence. During Rich semantic graph
generation module, a set of heuristic rules are applied to the
generated rich semantic graph to reduce it by merging,
deleting or consolidating the graph nodes.
Example Text
Fig. 4. Semantic graph text representation [13]
Fig 4.Semantic graph representation for the example
text represents the nodes in the rich semantic graph represents
the objects of the domain ontology classes for the input text
nouns and verbs.
Fig. 5. Semantic graph reduction summary [13]
Fig.5.represents the reduced semantic graph and the
summarized text obtained from the reduced graph. The
reduced semantic graph is obtained by applying reduction
heuristic rule. The generated abstractive summary contains
only 50 percent of the original text document.
Lloret.E et al. [14] developed a model for generating
ultra-concise concept-level summaries. The system initially
converts the input document into its syntactic representation
after lexical analysis. Summary generation is achieved using
language generation tool with the lexical units as the input.
This system lacks semantic representation of text and works
on an assumption that all the sentences are anaphora resolved.
Nikita Munot et al. [15] performed summarization
that the documents can be splitted into sentences and built by
SPO triples to create a semantic graph called rich semantic
graph(RSG). After performing RSG, it reduces the graph
nodes to Subject Noun(SN), Main Noun(MN), and Object
Noun(ON). After reducing Semantic graph, summary can be
generated.
Jurij Leskovec et al. [16] proposed that the original
document is represented as a semantic graph in the form of
triples consisting of subject-predicate-object. The sub-
structure of the graph is extracted to generate summaries.
SVM classifier is used to identify a set of triplet that
contribute to sentence extraction. A rich set of linguistic
attributes are incorporated into the model to increase the
performance of the proposed model. The set of triplets
generated in the previous step are refined through a series of
steps involving co-reference resolution, cross sentence
pronoun resolution and sentence normalization and finally
merge them to generate a semantic graph.

D. Semantic Text Representation Model
This technique aims to analyse input text using
ssemantics of words rather than syntax/Structure of text.
Khan Atif et al. [17] proposed a framework for
abstractive summarization of multiple documents in the form
of semantic representation of supply documents. Content
selection is done by ranking of the most significant predicate
argument structures. Finally summary is generated using a
language generation tool. But the system does not handle more
detailed semantics in the summarization approach due to its
assumption that the text to be handled are anaphora resolved
and sense disambiguated.
Atif et al. [18] suggested Semantic role labelling to
extract predicate argument structure from each sentence and
the document set is split into sentences with its document
number and position number. The position number is assigned
by using SENNA semantic role labeller API. The similarity
matrix is constructed from semantic Graph for Semantic
similarity scores. After that, modified graph based ranking
algorithm is used to determine predicate structure, semantic
similarity and document set relationship. After predicate,
MMR is used to reduce redundancy for summarization.
TABLE II
A COMPARATIVE STUDY ON SEMANTIC BASED APPROACH
Author/
Year
Techniques/
Methods
Text
Representation
Content
Selection
Summary
Generation
Albert
Gatt,
2009
Multimodal
Semantic
Based
Semantic model Simple
NLG
Generation
technique:
Synchronous
tree
Pierre-
Etienne
Genest,
2011
Information
item
Based
Abstract
representation:
Information
item(INIT)
Generated
sentences
can be
ranked
based on
document
frequency
NLG realize
simple NLG
Ibrahim,
2012
Semantic
Graph based Rich Semantic
graph
Calculation
of each
concepts
and
their
sentences
Reduced
Semantic
graph
Atif,
2015
Reduced
Semantic
graph
Semantic Role
Labelling
Semantic
graph for
similarity
score
NLG realize
simple NLG
IV .INFERENCES MADE
Quality of the summary is improved in structure
based approach since it produces coherent , less redundant
summary with higher coverage. The structure based method
may have some grammatical issues since it does not take
semantic representation of the document into consideration.
The semantic based model provides better linguistic quality to
the summary since it involves semantic representation of the
text document capturing the semantic relations. The semantic
method overcomes the issues of structure based that is it
reduces redundancy in the summary , ensures better cohesion
and also provides information rich content with better
linguistic quality.
V .CHALLENGES AND FUTURE RESEARCH
DIRECTIONS
The major issues of abstractive summarization is there
is no generalized framework , parsing and alignment of parse
trees is difficult. Extracting the important sentences and
sentence ordering as it has to be in the order as in the original
source document for producing an efficient summary is an
open issue. Compressions involving lexical substitutions,
paraphrase and reformulation is difficult with abstractive
summarization. The capability of the system is constrained by
the richness of their representation and their way to produce
such structure is the greatest challenge for abstractive
summary. Still Information diffusion is not handled properly
using abstractive text summarization.
VI. EXPERIMENTAL EVALUATION
Various benchmarking datasets are used for experimental
evaluation of abstractive summarization. Document
Understanding Conference(DUC) is the most common
benchmarking dataset used for text summarization. There are
number of datasets like DUC, TAC, DUC-2002,DUC-
2004,2005, CNN, DUC-2006, DUC-2007,2008, TIPSTER,
TREC. This dataset contains documents along with their
summaries that are created manually, automatically and
submitted summaries[19] [20] [21]
ROUGE toolkit is used to measure the summarization
performance, which is widely applied by DUC. Several
ROUGE methods are ROUGE-N, ROUGE-L, ROUGE-W and
ROUGESU. Higher order ROUGE-N score (N>1) estimates
the fluency of summaries.
VII. CONCLUSION
This survey has showcased various methods of abstractive
summarization. Abstractive summarization methods produce
highly cohesive, coherent, less redundant summary and
information rich. The goal is to provide an extensive survey
and comparison of different techniques and approaches of
abstractive summarization. In this survey some of the
challenges and future research directions are also highlighted.
In summary, the literature in abstractive summarization
depicts major progress in various aspects. However these

works still have not addressed the various challenges of
abstractive summarization to its full extent in terms of space
and time complexity.
REFERENCES
[1] R. Barzilay and K. R. McKeown, “Sentence fusion for multidocument
news summarization,” Comput. Linguist., vol. 31, no. 3, pp. 297–328,
Sep. 2005. [Online]. Available:
http://dx.doi.org/10.1162/089120105774321091
[2] S. H. Finley and S. M. Harabagiu, “Generating single and
multidocument summaries with gistexter,” in In U. Hahn & D. Harman
(Eds.), Proceedings of the workshop on automatic summarization, 2002,
pp.30–38.
[3] C.-S. Lee, Z.-W. Jian, and L.-K. Huang, “A fuzzy ontology and its
application to news summarization,” Systems, Man, and Cybernetics,
Part B: Cybernetics, IEEE Transactions on, vol. 35, no. 5, pp. 859–880,
Oct 2005.
[4] T. K. T. K. Hideki Tanaka, Akinori Kinoshita and N. Kato,
“Syntaxdriven sentence revision for broadcast news summarization,” in
NHK Science and Technology Research Labs. 1-10-11, Kinuta,
Setagaya-ku, Tokyo, Japan, 2009.
[5] P.-E. Genest and G. Lapalme, “Fully abstractive approach to guided
summarization,” in Proceedings of the 50th Annual Meeting of the
Association for Computational Linguistics: Short Papers - Volume 2,
ser. ACL ’12. Stroudsburg, PA, USA: Association for Computational
Linguistics, 2012, pp. 354–358. [Online]. Available:
http://dl.acm.org/citation.cfm?id=2390665.2390745
[6] H. T. Le and T. M. Le, “An approach to abstractive text summarization,”
in Soft Computing and Pattern Recognition (SoCPaR), 2013
International Conference of, Dec 2013, pp. 371–376.
[7] A. John and M. Wilscy, “Random forest classifier based multi-document
summarization system,” in Intelligent Computational Systems (RAICS),
2013 IEEE Recent Advances in, Dec 2013, pp. 31–36.
[8] D. Wang and T. Li, “Weighted consensus multi-document
summarization,” Information Processing & Management, vol. 48, no. 3,
pp. 513 – 523, 2012, soft Approaches to {IA} on the Web. [Online].
Available:
http://www.sciencedirect.com/science/article/pii/S0306457311000732
[9] K. Ganesan, C. Zhai, and J. Han, “Opinosis: A graph-based approach to
abstractive summarization of highly redundant opinions,” in Proceedings
of the 23rd International Conference on Computational Linguistics, ser.
COLING ’10. Stroudsburg, PA, USA: Association for Computational
Linguistics, 2010, pp. 340–348. [Online]. Available:
http://dl.acm.org/citation.cfm?id=1873781.1873820
[10] E. Lloret, D. o. S. Manuel Palomar, and A. S. Computing Systems
University of Alicante Apdo. de correos, 99 E-03080, “Analyzing the
use of word graphs for abstractive text summarization,” in The First
International Conference on Advances in Information Mining and
Management, 2011.
[11] A. Gatt and E. Reiter, “Simplenlg: A realisation engine for practical
applications,” in Proceedings of the 12th European Workshop on
Natural Language Generation, ser. ENLG ’09. Stroudsburg, PA, USA:
Association for Computational Linguistics, 2009, pp. 90–93. [Online].
Available: http://dl.acm.org/citation.cfm?id=1610195.1610208
[12] P.-E. Genest and G. Lapalme, “Framework for abstractive
summarization using text-to-text generation,” in Proceedings of the
Workshop on Monolingual Text-To-Text Generation, ser. MTTG ’11.
Stroudsburg, PA, USA: Association for Computational Linguistics,
2011, pp. 64–73.
[Online]. Available: http://dl.acm.org/citation.cfm?id=2107679.2107687
[13] [13] I. Moawad and M. Aref, “Semantic graph reduction approach for
abstractive text summarization,” in Computer Engineering Systems
(ICCES), 2012 Seventh International Conference on, Nov 2012, pp.
132– 138.
[14] E. Lloret, E. Boldrini, T. Vodolazova, P. MartÃ nez-Barco, R. MuÃ´soz,
and M. Palomar, “A novel concept-level approach for ultraconcise
opinion summarization,” Expert Systems with Applications, vol. 42, no.
20, pp. 7148 – 7156, 2015. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S0957417415003541
[15] N. Munot and P.-N. P. I. Sharvari S. Govilkar Department of Computer
Engineering, Mumbai University, “Conceptual framework for
abstractive text summarization,” International Journal on Natural
Language Computing (IJNLC) Vol. 4, No.1, February 2015, 2015.
[16] J. Leskovec, M. Grobelnik, and N. Milic-Frayling, “Learning
substructures of document semantic graphs for document
summarization,” in Proceedings of the KDD 2004 Workshop on Link
Analysis and Group Detection (LinkKDD), Seattle, WA, USA, 2004.
[17] A. Khan, N. Salim, and Y. Jaya Kumar, “A framework for
multidocument abstractive summarization based on semantic role
labelling,” Appl. Soft Comput., vol. 30, no. C, pp. 737–747, May 2015.
[Online]. Available: http://dx.doi.org/10.1016/j.asoc.2015.01.070
[18] A. Khan, N. Salim, and Y. Kumar, “Genetic semantic graph approach
for multi-document abstractive summarization,” in Digital Information
Processing and Communications (ICDIPC), 2015 Fifth International
Conference on, Oct 2015, pp. 173–181.
[19] http://duc.nist.gov/data.html.
[20] http://trec.nist.gov/data.html.
[21] http://www.nist.gov/tac/data/forms/.
View publication stats
View publication stats

AbstractiveSurvey of text in today timef

More Related Content

Similar to AbstractiveSurvey of text in today timef (20)

Recently uploaded (20)

AbstractiveSurvey of text in today timef