关键词: 定性概率网, 定性影响, 粗糙集, 概率下近似, 属性依赖度, 属性相对必要性

Abstract: Qualitative Probabilistic Network (QPN) is a powerful knowledge representation tool. However, sub-QPN can only represent sub-domain knowledge. To build a large QPN to represent the whole domain knowledge, an integration method of multiple sub-QPNs that have different nodes was proposed based on rough sets. Specifically, a single variable or a combination of multiple variables in a QPN could be regarded as an attribute in rough sets. First, multiple sub-QPNs were combined into an initial integrated QPN during integrating, then the directed edges and qualitative signs were added into the QPN according to attribute dependency degree, and then some unnecessary edges of which child node had multiple parent nodes could be deleted according to relative necessity of attribute. Thus, a large integrated QPN would be obtained to represent the whole domain knowledge. Finally, the experimental results illustrate that the integration method is feasible and effective.

Key words: Qualitative Probabilistic Network (QPN), qualitative influence, rough set, probabilistic lower approximation, attribute dependency degree, relative necessity of attribute