Sorry, you need to enable JavaScript to visit this website.

BNU: A BALANCE-NORMALIZATION-UNCERTAINTY MODEL FOR INCREMENTAL EVENT DETECTION

Citation Author(s):
Yunyan Zhang
Submitted by:
Jia Li
Last updated:
7 May 2022 - 12:13pm
Document Type:
Poster
Document Year:
2022
Event:
Presenters:
Jia Li
Paper Code:
SU3.E.4
Categories:
Keywords:

Abstract

Event detection is challenging in real-world application since new events continually occur and old events still exist which may result in repeated labeling for old events. There- fore, incremental event detection is essential where a model continuously learns new events and meanwhile prevents per- formance from degrading on old events. Although existing incremental event detection models achieve impressive per- formance, they face the data imbalance problem between old classes and new classes, and have the knowledge transfer prob- lem which cannot adequately utilize the knowledge provided by the previous model and data. To this end, we propose a Balance-Normalization-Uncertainty (BNU) model to address above problems. Specifically, in order to mitigate the adverse effects of data imbalance, we incorporate a balanced fine- tuning stage and a cosine normalization module. Meanwhile, we consider aleatoric uncertainty to preserve previous knowl- edge while training for new events. Experimental results show that our proposed method resolves the above challenges effec- tively and achieves consistent and significant performance on ACE and TAC KBP datasets.

up
0 users have voted:

Files

ICASSP22_3563.pdf

(38)