On the construction of a probabilistic model for assessing users' beliefs in an information environment

Abstract

We are developing an interaction environment for information portals in which users can create their own personal knowledge bases to store and define relationships between retrieved items. In this paper, we present our approach for combining these personal knowledge bases to a knowledge base for the entire user community that is to be used as part of an automatic recommendation system. We utilize a probabilistic model for assessing how strongly a single user's definition (or lack thereof) of a relationship should influence the system's belief in the relationship. The latter is revised using Pearl's method of uncertain evidence.