Abstract
Updating knowledge by the latest information has been treated in Bayesian probability models. Bayesian probability models, however, cannot express the ignorance reasonably due to the strict condition of probability. Owing to the flexibility of the model, the ignorance can be represented reasonably in Dempster-Shafer theory of evidence. In this paper, as a basis for modeling human knowledge updating process with information, the conventional combination rules of belief functions are examined. The changes of information measures of nonspecificity and dissonance with the update are investigated. A new combination rule having different properties is proposed.
