Event Detection in the Socio Political Domain

Emmanuel Cartier, Hristo Tanev


Abstract
In this paper we present two approaches for detection of socio political events: the first is based on manually crafted keyword combinations and the second one is based on a BERT classifier. We compare the performance of the two systems on a dataset of socio-political events. Interestingly, the systems demonstrate complementary performance: both showing their best accuracy on non overlapping sets of event types. In the evaluation section we provide insights on the effect of taxonomy mapping on the event detection evaluation. We also review in the related work section the most important resources and approaches for event extraction in the recent years.
Anthology ID:
2024.politicalnlp-1.2
Volume:
Proceedings of the Second Workshop on Natural Language Processing for Political Sciences @ LREC-COLING 2024
Month:
May
Year:
2024
Address:
Torino, Italia
Editors:
Haithem Afli, Houda Bouamor, Cristina Blasi Casagran, Sahar Ghannay
Venues:
PoliticalNLP | WS
SIG:
Publisher:
ELRA and ICCL
Note:
Pages:
12–21
Language:
URL:
https://aclanthology.org/2024.politicalnlp-1.2
DOI:
Bibkey:
Cite (ACL):
Emmanuel Cartier and Hristo Tanev. 2024. Event Detection in the Socio Political Domain. In Proceedings of the Second Workshop on Natural Language Processing for Political Sciences @ LREC-COLING 2024, pages 12–21, Torino, Italia. ELRA and ICCL.
Cite (Informal):
Event Detection in the Socio Political Domain (Cartier & Tanev, PoliticalNLP-WS 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.politicalnlp-1.2.pdf