基于粗糙集的定性概率网推理冲突解决方法

定性概率是贝叶斯网的定性抽象，它以有向边上的定性影响代替贝叶斯网中的条件概率参数，描述了变量间增减的趋势，具有高效的推理机制。但定性概率网中信息丢失导致推理的过程中往往产生不确定信息，即推理结果产生冲突。以尽可能消除定性推理中的冲突为出发点，在构建定性概率网时，基于粗糙集属性依赖度理论求解出网中节点间的依赖度，以依赖度作为变量间定性影响的权重，并根据依赖度改进已有的定性概率网推理算法，从而解决定性概率网推理冲突。实例验证表明，该方法既保持了定性概率网高效推理的特性，又能有效解决冲突。

Rough-set based approach to solve the inference conflict in qualitative probabilistic network

Qualitative Probabilistic Networks (QPNs) are the qualitative abstraction of Bayesian networks by substituting the conditional probabilistic parameters by qualitative influences on directed edges. Efficient algorithms have been developed for QPN reasoning. Due to the high abstraction, unresolved trade-offs (i.e., conflicts) during inferences with qualitative probabilistic networks may be produced. Motivated by avoiding the conflicts of QPN reasoning, a rough-set-theory based approach was proposed. The attribute association degrees between node peers were calculated based on the rough-set-theory while the QPNs were constructed. The association degrees were adopted as the weights to solve the conflicts during QPN inferences. Accordingly, the algorithm of QPN reasoning was improved by incorporating the attribute association degrees. By applying this method, the efficiency of QPNs inferences can be preserved, and the inference conflict can be well addressed at the same time.