基于案例和规则推理的贝叶斯网建模

Bayesian network modeling based on CBR and RBR

正确、高效地针对问题建立模型是应用贝叶斯网的关键,而从数据中学习贝叶斯网往往因为搜索空间庞大而效率低下.提出基于案例和规则推理的建模方法,建立领域知识库,使用框架和一阶概率逻辑表示贝叶斯网,当面对新的问题时,使用相似度和偏离度两个指标进行案例匹配,对选中的案例使用组合和剪枝技术修正,得到新问题的求解模型.整个过程以案例推理为主,并用规则推理辅助.这种方法能够复用历史案例,提高贝叶斯网建模效率.

The key of using Bayesian network is to correctly and efficiently construct models for problems. But, learning Bayesian network from data may be time expensive because of huge search space. In this paper, a modeling method based on case-based reasoning and rule-based reasoning is proposed. We build a domain knowledge base and represent Bayesian networks by frame and first-order probability logic. When facing a new problem, we use similarity ratio and difference ratio to match cases, and then combine and prune candidate cases to form a new model. In the whole process, case-based reasoning is the main technique, and rule-based reasoning plays an assistant role. This method directly reuses historical cases so as to improve Bayesian network modeling efficiency.