Temporal Relationship Extraction of Conflict Events in Open Source Military Journalism

Military conflict is a global concern, and as global conflicts continue to grow in number and complexity, the need to quickly and accurately understand the temporal relationships of conflict events is becoming more pressing. Analyses of the course of conflict events by military analysts, intelligence officers are crucial in war prevention, intervention and conflict resolution. Extracting event timing relationships from a large number of conflict-related texts in open source journalism can help researchers and policy makers better understand the evolution of conflicts and the causal relationships behind them. However, there are fewer current studies on event temporal relationship extraction from Chinese texts, while the accuracy of current temporal relationship studies is low. In this paper, we construct a chapter-level Chinese military conflict event temporal relationship dataset. In order to extract the event relationships in the document, the whole document is coded using Roberta, contextual extraction of multi-featured Chinese military conflict event trigger words is performed with additional classification headers, and contextual repetitions of different event pairs corresponding to the location of event triggers are sent as inputs, and then the model is fine-tuned. classification of event pairs labels using a linear layer. This method is more effective for the task of event temporal relationship extraction. The text constructs a Chinese military event conflict event dataset, and the experimental results show that the recall F1 based on this method reaches 0.63, which is slightly higher than the current mainstream benchmark model.