从确信因子模型到Bayes网络

From Certainty Factor Model to Bayesian Network

本文研究确信因子模型与Bayes网络之间的区别与联系。首先讨论确信因子模型理论基础的局限性,证明确信因子模型中蕴含着与简单Bayes模型一样的条件独立性假设;然后探究Bayes网络中对应于确信因子模型的若干功能,提出Bayes网络推理中条件对推理结论的影响程度与作用方向的概念、分析方法和计算公式,证明Noisy-OR模型的概率推理与确信因子的推理的等价性;最后从知识的表示、推理、获取等三个方面讨论Bayes网络相对于确信因子模型的比较优势。本文的研究表明Bayes网络不仅具备确信因子模型的主要功能,而且可以突破确信因子的局限性。它有望取代确信因子模型,成为基于概率的智能信息处理模型中的一种主流模型。

In this paper, the relations and differences between certainty factor model and Bayesian Network are researched. Firstly, the limitation of the theoretical basis of certainty factor model is discussed; it is proved that the certainty factor model implies a conditional independence hypothesis same as the simple Bayesian model implies. Then, some functions of Bayesian Network, which correspond to certainty factor model, are explored. The concepts, analysis approach, and computing formula of condition's influence degree and effect direction to the inference conclusion in Bayesian network are presented, and it is also proved that the equivalence of the probabilistic inference between the Noisy-OR model and certainty factor model. Finally, the superiority of Bayesian network to certainty factor model is discussed in the aspects of representation, inference, and acquire of the knowledge. Our conclusion is that the Bayesian network not only has the main function of the certainty factor model, but also can break through some limitation of this model, so the Bayesian network can be a main trend probabilistic model in intelligent information process instead of the certainty factor model eventually.