ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2016, Vol. 53 ›› Issue (7): 1544-1560.doi: 10.7544/issn1000-1239.2016.20148251

• 人工智能 • 上一篇    下一篇



  1. 1(国防科学技术大学计算机学院 长沙 410073); 2(国防科学技术大学人文与社会科学学院 长沙 410073) (
  • 出版日期: 2016-07-01
  • 基金资助: 

ICIC_Target: A Novel Discovery Algorithm for Local Causality Network of Target Variable

Li Yan1, Wang Ting1, Liu Wanwei1, Zhang Xiaoyan2   

  1. 1(College of Computer, National University of Defense Technology, Changsha 410073);2(College of Humanities and Social Sciences, National University of Defense Technology, Changsha 410073)
  • Online: 2016-07-01

摘要: 因果关系的研究在于揭示自然规律的和人类社会发展本质及其规律,对人类长久以来的生产生活和科学研究有着非常重要的作用.目前,因果关系的研究受到前所未有的广泛关注,但仍存在诸多困难和挑战.致力于建立一个因果激励抑制模型以抽象地表示和解释因果的作用机制,并在此基础上提出用于目标节点的局部因果关系网络的自动发现方法框架ICIC和算法ICIC_Target.该方法不预先设定因果结构(如设定为无圈、隐含结构),并根据对因果关系本质的认识,利用初始变量(exogenous variables)和初始团树(IClique)的概念,在判定边和方向之前对变量进行粗略地排序,从而提高了因果关系网络发现的性能.在4个不同类型的数据集上实现了与多种经典方法,如HITON,IC,PC,PCMB等的对比实验,实验结果表明ICIC_Target方法适用范围广,有较好的鲁棒性,同时,从理论上分析证实了ICIC_Target方法具有较好的稳定性和较低的复杂度.

关键词: 因果关系, 因果关系网络, 局部因果关系发现, 激励因果, 抑制因果

Abstract: Causality research aims to reveal the law of evolution of nature, society and human. Nowadays, the causality research receives widespread attention for its important applications of human life and science research, but there are still many difficulties and challenges. This paper presents a unified model to explain the stimulating and inhibiting causalities. Based on this model, we also present a framework ICIC and a novel algorithm ICIC_Target to infer the local causal structure of a target variable from observational data without any limitation of some assumptions, such as assumption of acyclic structure, hidden variables and so on. Following our descriptions of causality essence and properties, as well as several classical theories proposed by Judea Pearl, Gregory F. Cooper and so on, we introduce concepts of exogenous variable and clique-like structure (IClique) to get rough ordering of variables, which are necessary for revealing the causality accurately and efficiently. To evaluate our approach, several experiments compared with HITON, IC, PC, PCMB and several methods based on four datasets with different data types have been done. The results demonstrate the higher performance and stronger robustness of our algorithm ICIC_Target. In this paper, we also discuss the advantages of stability and complexity of ICIC_Target.

Key words: causality, causal network, local causal network discovery, stimulating causality, inhibiting causality