A summary of any event type is only complete if certain information aspects are mentioned. For a court trial, readers will at least want to know who is involved and what the charges and the sentence are. For a natural disaster, they will ask for the disaster type, the victims and other damages. Will a co-occurrence or frequency-based sentence extraction summariser automatically provide the requested information, or are the results better if an information extraction (IE) system first detects the summary-crucial aspects? To answer this question, we compared the performance of a purely co-occurrence-based method with a system that additionally makes use of targeted IE. As each event type requires different information aspects and not all of them were covered by the existing IE software, we used a tool that learns semantically related terms to cover the remaining aspects. The comprehensive evaluation in the TAC'2010 competition showed that event extraction is indeed beneficial for summarisation performance, and that summary quality is directly related to
IE quality. Our integrated system was ranked among the top systems participating at TAC.
STEINBERGER Josef;
TANEV Hristo;
KABADJOV Mijail;
STEINBERGER Ralf;
2013-06-06
Bahri Publications
JRC78738
0976-0962,
http://bahripublications.co.in/journal.php?cat_id=31,
https://pdfs.semanticscholar.org/fb40/8e1cf228fff48ecdf8131cef503190c516f1.pdf,
https://publications.jrc.ec.europa.eu/repository/handle/JRC78738,
| Name | Country | City | Type |
|---|
This document is only visible at the Commission level.
You are not authorized to publish or distribute it outside the European Commission.
This is a public document. You can share this publication.
Datasets
| ID | Title | Public URL |
|---|
Dataset collections
| ID | Acronym | Title | Public URL |
|---|
Scripts / source codes
| Description | Public URL |
|---|
Additional supporting files
| File name | Description | File type |
|---|