event-centric summary generation lucy vanderwende, michele banko and arul menezes one microsoft way,...
TRANSCRIPT
![Page 1: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/1.jpg)
Event-Centric Summary Generation
Lucy Vanderwende, Michele Banko and Arul Menezes
One Microsoft Way, WA, USA DUC 2004
![Page 2: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/2.jpg)
2
Abstract
• Our primary interest is two folds:– To explore an event-centric approach to
summarization– To explore a generation approach to summary
realization
![Page 3: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/3.jpg)
3
Introduction
• Identifying important events, as opposed to entities
• Generation component– Human-authored rely less on sentence
extraction
• Graph-scoring algorithm– To identify highest weighted node to guide
content selection
![Page 4: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/4.jpg)
4
System Description
• MSR-NLP– Analysis component
• Rule-base syntactic analysis component• Produces a logical form
– Syntactic variations, words label
– Generation component• Syntactic realization component• Produces a syntactic tree
![Page 5: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/5.jpg)
5
Creating document representations
• Cluster sentence
• Analysis sentence and get logical form
![Page 6: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/6.jpg)
6
Creating document representations
• Produces triples result from logical form– (LFNodei, rel, LFNodej)
![Page 7: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/7.jpg)
7
Forming Document Graph
• Take those triples and join nodes by way of their semantic relation using a bidirectional link structure
• Keep track of how many times we observe the relationship
• Stop words are not included in the graph construction
![Page 8: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/8.jpg)
8
![Page 9: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/9.jpg)
9
Node scoring Using Pagerank
• Using Pagerank algorithm– Hyperlink such as WWW– When link between nodes, vote for that node
–
![Page 10: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/10.jpg)
10
Node scoring Using Pagerank
• Pagerank framework– “Pages”, correspond to base forms of words in the do
cuments– “hyperlink”, correspond to semantic relationships– Verbs, identify events– Noun, Identify entities– Use event to identify summary content
• Typically, the algorithm converges around 40 iterations
![Page 11: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/11.jpg)
11
Graph Scoring
• Use pagerank scores to assess the link weight (LW(i->n))
•
![Page 12: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/12.jpg)
12
Summary Generation
• Generated by extracting and merging of logical form– Identify important triples
• Defined highly link weight node, and together with most highly weighted
• (leave, Tobj, LonLondon_Bridge_Hospital)• Not (leave, Tobj, government)
– Extract fragments divided into “event” and “entity”
• Event used to generate summary• Entity used to expanded upon reference to the sa
me entity within the selected event fragment
![Page 13: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/13.jpg)
13
![Page 14: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/14.jpg)
14
Summary Generation
• Event fragment order– Cluster event fragment by they refer to – Choose the greatest number of argument nod
e for the event– Order the selected event fragments
• To group sentence referring to the same entity together
• Order sentence which exhibit event-coreference
![Page 15: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/15.jpg)
15
Experiments and Evaluation
•
(Rule-based pronoun resolution method, 75% accuracy)
![Page 16: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/16.jpg)
16
Experiments and Evaluation
•
Reason: the potential to introduce disfluent text
![Page 17: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/17.jpg)
17
![Page 18: Event-Centric Summary Generation Lucy Vanderwende, Michele Banko and Arul Menezes One Microsoft Way, WA, USA DUC 2004](https://reader035.vdocuments.site/reader035/viewer/2022062408/56649f295503460f94c42778/html5/thumbnails/18.jpg)
18
Directions and Future Work
• Produce more human-like generated summaries
• Further study the impact of anaphora resolution
• Study new page-ranking algorithm• While ordering groups event fragments
mentioning the same entity, we have not yet implemented a system to combine them into larger logical form construction