Details
| Originalsprache | Englisch |
|---|---|
| Titel des Sammelwerks | WWW '23 |
| Untertitel | Proceedings of the ACM Web Conference 2023 |
| Seiten | 3959-3967 |
| Seitenumfang | 9 |
| ISBN (elektronisch) | 9781450394161 |
| Publikationsstatus | Veröffentlicht - 30 Apr. 2023 |
| Veranstaltung | 2023 World Wide Web Conference, WWW 2023 - Austin, USA / Vereinigte Staaten Dauer: 30 Apr. 2023 → 4 Mai 2023 |
Abstract
The recent widespread use of social media platforms has created convenient ways to obtain and spread up-to-date information during crisis events such as disasters. Time-critical analysis of crisis data can help human organizations gain actionable information and plan for aid responses. Many existing studies have proposed methods to identify informative messages and categorize them into different humanitarian classes. Advanced neural network architectures tend to achieve state-of-the-art performance, but the model decisions are opaque. While attention heatmaps show insights into the model's prediction, some studies found that standard attention does not provide meaningful explanations. Alternatively, recent works proposed interpretable approaches for the classification of crisis events that rely on human rationales to train and extract short snippets as explanations. However, the rationale annotations are not always available, especially in real-time situations for new tasks and events. In this paper, we propose a two-stage approach to learn the rationales under minimal human supervision and derive faithful machine attention. Extensive experiments over four crisis events show that our model is able to obtain better or comparable classification performance (∼86% Macro-F1) to baselines and faithful attention heatmaps using only 40-50% human-level supervision. Further, we employ a zero-shot learning setup to detect actionable tweets along with actionable word snippets as rationales.
ASJC Scopus Sachgebiete
- Informatik (insg.)
- Computernetzwerke und -kommunikation
- Informatik (insg.)
- Software
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
WWW '23: Proceedings of the ACM Web Conference 2023. 2023. S. 3959-3967.
Publikation: Beitrag in Buch/Bericht/Sammelwerk/Konferenzband › Aufsatz in Konferenzband › Forschung › Peer-Review
}
TY - GEN
T1 - Learning Faithful Attention for Interpretable Classification of Crisis-Related Microblogs under Constrained Human Budget
AU - Nguyen, Thi Huyen
AU - Rudra, Koustav
N1 - Publisher Copyright: © 2023 ACM.
PY - 2023/4/30
Y1 - 2023/4/30
N2 - The recent widespread use of social media platforms has created convenient ways to obtain and spread up-to-date information during crisis events such as disasters. Time-critical analysis of crisis data can help human organizations gain actionable information and plan for aid responses. Many existing studies have proposed methods to identify informative messages and categorize them into different humanitarian classes. Advanced neural network architectures tend to achieve state-of-the-art performance, but the model decisions are opaque. While attention heatmaps show insights into the model's prediction, some studies found that standard attention does not provide meaningful explanations. Alternatively, recent works proposed interpretable approaches for the classification of crisis events that rely on human rationales to train and extract short snippets as explanations. However, the rationale annotations are not always available, especially in real-time situations for new tasks and events. In this paper, we propose a two-stage approach to learn the rationales under minimal human supervision and derive faithful machine attention. Extensive experiments over four crisis events show that our model is able to obtain better or comparable classification performance (∼86% Macro-F1) to baselines and faithful attention heatmaps using only 40-50% human-level supervision. Further, we employ a zero-shot learning setup to detect actionable tweets along with actionable word snippets as rationales.
AB - The recent widespread use of social media platforms has created convenient ways to obtain and spread up-to-date information during crisis events such as disasters. Time-critical analysis of crisis data can help human organizations gain actionable information and plan for aid responses. Many existing studies have proposed methods to identify informative messages and categorize them into different humanitarian classes. Advanced neural network architectures tend to achieve state-of-the-art performance, but the model decisions are opaque. While attention heatmaps show insights into the model's prediction, some studies found that standard attention does not provide meaningful explanations. Alternatively, recent works proposed interpretable approaches for the classification of crisis events that rely on human rationales to train and extract short snippets as explanations. However, the rationale annotations are not always available, especially in real-time situations for new tasks and events. In this paper, we propose a two-stage approach to learn the rationales under minimal human supervision and derive faithful machine attention. Extensive experiments over four crisis events show that our model is able to obtain better or comparable classification performance (∼86% Macro-F1) to baselines and faithful attention heatmaps using only 40-50% human-level supervision. Further, we employ a zero-shot learning setup to detect actionable tweets along with actionable word snippets as rationales.
KW - Classification
KW - Crisis Events
KW - Interpretability
KW - Twitter
UR - http://www.scopus.com/inward/record.url?scp=85159375858&partnerID=8YFLogxK
U2 - 10.1145/3543507.3583861
DO - 10.1145/3543507.3583861
M3 - Conference contribution
AN - SCOPUS:85159375858
SP - 3959
EP - 3967
BT - WWW '23
T2 - 2023 World Wide Web Conference, WWW 2023
Y2 - 30 April 2023 through 4 May 2023
ER -