Bayesian Networks are probabilistic structured representations of domains which have been applied to monitoring and manipulating cause and effects for modelled systems as disparate as the weather, disease and mobile telecommunications networks. Although useful, Bayesian Networks are notoriously difficult to build accurately and efficiently which has somewhat limited their application to real world problems. Ontologies are also a structured representation of knowledge, encoding facts and rules about a given domain. This paper outlines an approach to harness the knowledge and inference capabilities inherent in an ontology model to automate the construction of Bayesian Networks to accurately represent a domain of interest. The approach was implemented in the context of an adaptive, self-configuring network management system in the telecommunications domain. In this system, the ontology model has the dual function of knowledge repository and facilitator of automated workflows and the generated BN serves to monitor effects of management activity, forming part of a feedback look for self-configuration decisions and tasks.
Two-way latent grouping model for user preference prediction
Eerika Savia, Kai Puolamäki, Janne Sinkkonen and Samuel Kaski
In: UAI 2005, 26-29 July 2005, Edinburgh, Scotland.
Fang Wu and Bernardo A. Huberman
HP Laboratories
Palo Alto, CA 94304
January 23, 2008
Abstract
We analyze the role that popularity and novelty play in attracting
the attention of users to dynamic websites. We do so by determining
the performance of three different strategies that can be utilized to
maximize attention. The first one prioritizes novelty while the second
emphasizes popularity. A third strategy looks myopically into
the future and prioritizes stories that are expected to generate the
most clicks within the next few minutes. We show that the first two
strategies should be selected on the basis of the rate of novelty decay,
while the third strategy performs sub-optimally in most cases. We also
demonstrate that the relative performance of the first two strategies
as a function of the rate of novelty decay changes abruptly around a
critical value, resembling a phase transition in the physical world.
N. Tatti, T. Mielikainen, A. Gionis, and H. Mannila. ICDM '06: Proceedings of the Sixth International Conference on Data Mining, page 603--612. Washington, DC, USA, IEEE Computer Society, (2006)