基于时序集成图谱和图嵌入的突发事件预警方法

Early Warning Method for Emergency Events Based on Temporal Knowledge Graph and Graph Embedding

针对当前突发事件预警研究在使用异构多源信息和形势快速变化情况下的局限性,结合时序集成图谱和图嵌入技术提出了一种基于时序集成图谱分割、静态图降维和目标函数拟合优化的突发事件预警方法。首先,构建了时序集成图谱,将采集的突发事件关联信息实时处理后,关联到分层的实体,进行图谱结构化;然后,利用图嵌入方法计算图谱节点的嵌入向量,根据目标函数初步计算突发事件威胁值,拟合历史事件发展趋势修正目标函数参数,实时计算事件最新预警结果;最后,通过试验验证了该方法的有效性。

Aimed at the limitations of the current research on early warning methods for the emergency events with heterogeneous multi-source information and the rapid-change situation,an early warning method for the emergency events combined with temporal knowledge graph and graph embedding technology is proposed,by cutting apart graph,static graph dimension reduction and optimizing the target function.Firstly,a temporal knowledge graph is constructed.After real-time processing collected information related with the emergency events,information is associated with layered entities,and the graph structure is formed.Then,with graph embedding method,the embedding vector of the graph node is calculated.The emergency threat value is preliminarily calculated according to the target function.The parameters of the target function are modified by fitting historical data.Thus,the newest early warning result is real-time calculated.Finally,the effectiveness of the method is verified by experiments.