摘要
科学构建观察现象背后因果关系是各领域研究的一个基本问题。oCSE是构建动态因果网络的一种经典方法,它通过依次寻找单个节点的因果父集逐层构建网络,区别于当前通用的由成对节点间因果关系简单合成网络的思路。oCSE能够更充分利用数据生成高质量网络,但其存在两点局限:当出现多重传递性或共因性因素时易误判因果关系;大量测算对比致使时间效率偏低。为克服上述局限,本研究提出一种改进方法oCSE-BS:引入贝叶斯评分推断特殊情况下测试节点与目标节点的因果关系,避免引入伪父节点,提升识别因果关系的正确率;采取早期丢弃策略过滤弱相关节点,避免完全搜索带来的高计算量,提升算法运行的时间效率。经验证oCSE-BS在生成网络质量和时间效率方面均优于oCSE,同时发现其运行效果对网络规模、网络稀疏度敏感度较高,对样本噪音敏感度较低。
Reconstructing the causal relations behind the phenomena we observe is a fundamental problem in all fields of science.As a new method,oCSE constructs the network layer by layer by finding the causal parent set of a single node in turn,which is different from the current general idea of simply synthesizing the network from the causal relationship between pairs of nodes.In order to overcome the above limitations,we proposed an improved method oCSE-BS:Bayesian score was introduced to infer the causal relationship between the test node and the target node in special cases,so as to avoid the introduction of pseudo parent node;early discarding strategy was adopted to filter the weak correlation nodes,so as to avoid the high computational burden caused by complete search.It is verified that ocse-bs is superior to oCSE in terms of network quality and time efficiency.At the same time,it is found that its operation effect is more sensitive to network size and network sparsity,and less sensitive to sample noise.
作者
韩梦瑶
李雯
陈克斌
HAN Mengyao;LI Wen;CHEN Kebin(Department of Defense Economics,Army Logistical Academy,Chongqing 400030,China;78156 Troop,Chongqing 400030,China;College of Information and Communication,National University of Defense Technology,Wuhan 430019,China)
出处
《湖北大学学报（自然科学版）》
CAS
2024年第3期392-401,共10页
Journal of Hubei University：Natural Science
基金
国家社科科学基金军事学青年项目(2022-SKJJ-C-015)资助。
关键词
传递熵
因果熵
贝叶斯评分
动态因果网络
transfer entropy
causation entropy
Bayesian score
dynamic causal network
