A dataset of domain events based on open-source military news

Abstract

The text dataset of the military field is the basis for event extraction in the military field, and the datasets of high quality can effectively promote the study of event extraction in this field. However, the event extraction dataset commonly used in the real world (such as ACE2005, etc.) is oriented to the general field, and the text corpus resources on military events are scarce. Therefore, we collected a large amount of military news content from public military news websites. Firstly, on the basis of text content analysis, we first established an event model of military news including event types, entity types and entity relationship types. Secondly, we manually labeled the text data according to the event model, which was iteratively verified and corrected simultaneously. Finally, we obtained dataset of 13,000 high-quality military news events with a full variety of labels. We make this military news event dataset publicly available in this paper.