论文部分内容阅读
针对符号传播算法在符号相反的两条平行路径上进行推理时常常产生歧义性,提出一种基于定性互信息的歧义性约简方法.首先,给出定性互信息的严格定义.然后,提出基于定性互信息影响强度的定性概率网,进一步区分影响强度,并证明具有强度的定性影响的对称性、传递性和复合性.最后在Antibiotics数据集上,通过与已有方法推理结果的对比实验,验证该歧义性约简方法的正确性和高效性.理论分析和实验结果表明,基于定性互信息的定性概率网既保留定性推理的简明性,又能够有效约简定性推理的歧义性.
Aiming at the ambiguity when the symbol propagation algorithm performs inference on two parallel paths with opposite signs, a disambiguation method based on qualitative mutual information is proposed. First, the strict definition of qualitative mutual information is given. Then, The qualitative probability network that qualitatively influences the intensity of mutual information, further distinguishes the intensity of influence, and proves the symmetry, transitivity and compound of the qualitative impact with intensity.Finally, on the Antibiotics dataset, by comparing with the existing methods to reason the results, The correctness and high efficiency of the disambiguation method are verified.The theoretical analysis and experimental results show that the qualitative probability network based on qualitative mutual information not only retains the conciseness of qualitative reasoning but also effectively reduces the ambiguity of qualitative reasoning.