Bayesian Metanetworks for Modelling User Preferences in Mobile Environment

The problem of profiling and filtering is important particularly for mobile information systems where wireless network traffic and mobile terminal’s size are limited comparing to the Internet access from the PC. Dealing with uncertainty in this area is crucial and many researchers apply various probabilistic models. The main challenge of this paper is the multilevel probabilistic model (the Bayesian Metanetwork), which is an extension of traditional Bayesian networks. The extra level(s) in the Metanetwork is used to select the appropriate substructure from the basic network level based on contextual features from user’s profile (e.g. user’s location). Two models of the Metanetwork are considered: CMetanetwork for managing conditional dependencies and R-Metanetwork for modelling feature selection. The Bayesian Metanetwork is considered as a useful tool to present the second order uncertainty and therefore to predict mobile user’s preferences.