#GENERAL
-
EACL 2006 (Rank A in NLP conference)
-
University of British Columbia Vancouver, Canada
-
First Author: Giuseppe Carenini, Associate Professor of Computer Science, University of British Columbia
-
Work series on same topic of Giuseppe Carenini
- 》》MULTI-DOCUMENT SUMMARIZATION OF EVALUATIVE TEXT, journal version, Journal of Computational Intelligence 2013
- 》》Extractive vs. NLG-based abstractive summarization of evaluative text: The effect of corpus controversiality, INLG 2008 (rank B/B2)
- 》》Multi-Document Summarization of Evaluative Text, EACL 2006 (rank A/A2)
- 》》Extracting knowledge from evaluative text, K-Cap 2005 (rank A/B1), 196 citations
-
Related papers from other authors
-
Modeling content and structure for abstractive review
summarization, Journal of Computer Speech and Language 2016 - Abstractive Summarization of Product Reviews Using Discourse Structure, EMNLP 2014, Giuseppe as second author
-
》》Building a Sentiment Summarizer
for Local Service Reviews, WWW-2008 workshop on NLP in the Information Explosion Era, Google Inc.
-
Modeling content and structure for abstractive review
#ABSTRACT & CONCLUSION
- compare two approaches of summarization
- sentence extraction based approach MEAD (an open source package)
- language generation based approach SEA
- conclusion
- both perform equally well quantitatively
- MEAD: varied language and details but lack in accuracy, fail to give an overview
- SEA: provide general overview but sounding 'robotic', repetitive and incoherent (rời rạc, ko mạch lạc)
- both perform different but for complementary reasons
- should synthesize two approaches
- both perform equally well quantitatively
#INTRODUCTION
-
INDUSTRIAL NEEDS
- Online customer reviews, summaries of this literature could be of great strategic value to product designers, planners and manufactures
- Other important commercial applications such as summarization of travel logs
- non-commercial applications such as the summarization of candidate reviews
-
PROBLEM
- how to effectively summarize a large corpora of evaluative text about a single entity e.g. a product
- for factual documents, the goal is to extract important facts and present them in a sensible ordering while avoiding repetition
- when documents contain inconsistent info e.g. conflicting report, the goal is to identify overlaps and inconsistencies and produce a summary that points out and explain those inconsistencies.