Study on Logical Argumentation and Bayesian Network Combination 

Author  HuXiaoFeng 
Tutor  XingYongKang 
School  Chongqing University 
Course  Computer Software and Theory 
Keywords  logical argumentation Bayesian network explanation mechanic effectcause graph breast cancer recurrence 
CLC  TP183 
Type  Master's thesis 
Year  2007 
Downloads  102 
Quotes  0 
Bayesian network model have been widely used in expert system of artificial intelligence, which combine logical causality with probabilistic computation to reduce the complexity of probability reasoning. On the one hand, Bayesian network model ensures the feasibility of inference. On the other hand, it also ensures a reliable explain for every result which deduced from Bayesian network. Bayesian network model create a classical and successful exemplification about combination of logic and probability. In spite of the reliability of Bayesian network model on probabilistic deduction, how could we find a reasonable and logical explanation for the inferential result? According to the Pearl who was the founder of Bayesian network, the network structure indicates a kind of logical causality, so the reasoning process is an argumentation process to the result. In the real application, the fact is not so, the network structure was built by experts who have in charge of professional knowledge on some certain area, usually, the initial network was not a suitable network satisfied with the presupposition of Bayesian model, but a casualeffect graph. The initial graph denotes logical and real causeeffect relationships which express human expert’s professional knowledge. Because that the effect nodes in causeeffect graph have more than two cause nodes commonly, it will take no effective influence on Bayesian reasoning. So in order to utilize the wellrounded inbeing Bayesian network reasoning technology, the causeeffect must be transformed. Although the result of transformation may cause a satisfied Bayesian network, but at the same time another problem come about, that is the initial causeeffect graph must be changed, and the Bayesian network after transformation would not denote a complete and logical cause and effect relationship expressed the professional knowledge of human experts. Halfbaked information will attack the assumption on Pearl’s about one reasoning process is an argumentation process. If not take Bayesian network as argument on the explanation of inferential result, but obtain complete arguments from causeeffect graph, and find suitable arguments for the result. The process about finding causes for the being result just like searching arguments for or against the result, this kind of thinking model is similar with the logical argumentation system. So the main goal is combine the argumentation model which belongs to logic area and the Bayesian network model which belongs to probability area, to do so, it would satisfied not only utilizing mature Bayesian network reasoning to compute a precise probability value ,but also having a good and reliable explanation for the value.