en-es
en
0.25
0.5
0.75
1.25
1.5
1.75
2
HEADY: News Headline Abstraction Through Event Pattern Clustering
Published on Oct 02, 20132660 Views
This paper presents HEADY: a novel, ab- stractive approach for headline generation from news collections. From a web-scale corpus of English news, we mine syntactic patterns that a Noisy-OR model gene
Related categories
Chapter list
News Understanding Abstraction through Event Pattern Clustering00:00
From selection to generation00:41
Roadmap to abstractive headline generation01:08
Agenda - 101:50
Agenda - 202:05
Event Relatedness in News Collections - 102:07
Event Relatedness in News Collections - 202:45
Patterns and news collections - 103:48
Patterns and news collections - 204:45
Learning abstract representations of events (I)04:57
Learning abstract representations of events (II)06:43
Pattern extraction in detail07:46
Disambiguating entities (II)09:57
Co-occurrence and equivalence11:46
A glimpse inside the model - 112:26
A glimpse inside the model - 212:55
A glimpse inside the model - 313:18
Agenda - 313:42
Abstractive headline generation13:44
Headline examples14:42
Agenda - 415:06
Evaluation of headline generation - 115:09
Evaluation of headline generation - 216:53
Evaluation of headline generation - 317:13
Evaluation of headline generation - 418:08
Evaluation of headline generation - 518:39
Evaluation of headline generation - 619:02
Agenda - 519:42
Conclusions19:45
Directions for improvement20:28