非参数触发算法对时空因果关系的优化研究

Optimization of the spatio-temporal causal relationships by non-parametric triggering algorithm

为了从多类型时空序列中稳定准确地提取事件类型间的因果关系,提出了一种非参数触发算法(NPTP).首先基于多变量Hawkes模型,用非参数条件强度函数约定出事件类型间的因果关系;然后通过迭代计算事件类型间触发关系的条件概率;最后由秩选择得到显著事件类型对的概率并将概率中值的均值作为显著性度量值,从而获得事件类型间的触发关系.仿真结果表明,与参数算法CSTP和单变量算法MISD相比,在算法的稳定性和准确性指标方面均有优化提升.

In order to stably and accurately extract the causal relationship between event types from multi-type spatio-temporal sequences, in this paper, we propose a non-parametric triggering pattern algorithm(NPTP). Firstly,based on the multivariate Hawkes model, the non-parametric conditional intensity function is used to agree on the causal relationship between event types. The conditional probabilities of the triggering relationships between event types are calculated by iterative process afterwards. After the probabilities of the distinct pairwise event types are obtained by rank selection. Finally, the mean of the median values of the probabilities is used as the significance measure to obtain the triggering relationship between the event types. The simulation results show that compared with the parameter algorithm CSTP and the single variable algorithm MISD, the stability and accuracy of the NPTP algorithm are optimized.