Ilmu Sosial editorial
Open AccessOA2024
Socio-political Events of Conflict and Unrest: A Survey of Available Datasets
This survey maps the landscape of datasets for socio-political conflict and unrest, highlighting divergent annotation approaches between NLP and political science. It offers a unified view to foster cross-disciplinary research.
H. Olsen; Étienne Simon; Erik Velldal; Lilja Øvrelid· CASE· 2024· DOI 10.18653/v1/2024.case-1.5
The core problem
The study of socio-political events of conflict and unrest relies on a diverse array of datasets, each shaped by the disciplinary backgrounds of their creators. Scholars in natural language processing (NLP) typically annotate specific text spans that denote event components, while political scientists and conflict studies researchers often code abstract, structured representations of events that are less tied to source text. This survey aims to map the current landscape of event datasets across both communities, providing a unified view of the work done in different disciplines. The authors—H. Olsen, Étienne Simon, Erik Velldal, and Lilja Øvrelid—present their findings in the Proceedings of the CASE workshop, highlighting the need for integration.
Innovation
The survey methodology involves a systematic review of existing datasets within the domain of social and political conflict and unrest. The authors examine datasets from both NLP and political science communities, analyzing their creation approaches, annotation schemes, and intended research applications. The paper categorizes datasets based on their focus: NLP datasets emphasize span-level annotations tied to text, while political science datasets provide structured event codings that abstract away from specific texts. This comparative analysis reveals the strengths and limitations of each approach and identifies gaps in the current data landscape.
Introduction
The study of socio-political events of conflict and unrest relies on a diverse array of datasets, each shaped by the disciplinary backgrounds of their creators. Scholars in natural language processing (NLP) typically annotate specific text spans that denote event components, while political scientists and conflict studies researchers often code abstract, structured representations of events that are less tied to source text. This survey aims to map the current landscape of event datasets across both communities, providing a unified view of the work done in different disciplines. The authors—H. Olsen, Étienne Simon, Erik Velldal, and Lilja Øvrelid—present their findings in the Proceedings of the CASE workshop, highlighting the need for integration.
Why it matters
The discussion highlights the complementary nature of NLP and political science approaches. NLP datasets provide rich textual grounding and detailed event components, which can enhance the granularity of political event data. Conversely, political science datasets offer structured, scalable representations that can inform NLP models about event schemas and real-world complexities. The authors argue for greater collaboration between disciplines to create datasets that combine the strengths of both: text-anchored annotations with structured event coding. They also emphasize the need for standardized evaluation metrics and shared ontologies to facilitate dataset integration. Such efforts would enable more comprehensive studies of conflict and unrest, leveraging both computational and social science perspectives.
Who should read this
Pembaca Research Digest dan praktisi rumpun ini.
Opening member content…